From 32 to 2 ports: Ideal SATA/SAS Controllers for ZFS & Linux MD RAID

[Last updated on 2014-06-26]

I need a lot of reliable and cheap storage space (media collection, backups). Hardware RAID tends to be expensive and clunky. I recognize quite a few advantages in ZFS on Solaris/FreeBSD, and Linux MD RAID:

  • Performance. In many cases they are as fast as hardware RAID, and sometimes faster because the OS is aware of the RAID layout and can optimize I/O patterns for it. Indeed, even the most compute intensive RAID5 or 6 parity calculations take negligible CPU time on a modern processor. For a concrete example, Linux 2.6.32 on a Phenom II X4 945 3.0GHz computes RAID6 parity at close to 8 GB/s on a single core (check dmesg: "raid6: using algorithm sse2x4 (7976 MB/s)"). So achieving a throughput of 500 MB/s on a Linux MD raid6 array requires spending less than 1.5% CPU time computing parity. Now regarding the optimized I/O patterns, here is an interesting anecdote: one of the steps that Youtube took in its early days to scale their infrastructure up was to switch from hardware RAID to software RAID on their database server. They noticed a 20-30% increase in I/O throughput. Watch Seattle Conference on Scalability: YouTube Scalability @ 34'50".
  • Scalability. ZFS and Linux MD RAID allow building arrays across multiple disk controllers, or multiple SAN devices, alleviating throughput bottlenecks that can arise on PCIe links, or GbE links. Whereas hardware RAID is restricted to a single controller, with no room for expansion.
  • Reliability. No hardware RAID = one less hardware component that can fail.
  • Ease of recoverability. The data can be recovered by putting the disks in any server. There is no reliance on a particular model of RAID controller.
  • Flexibility. It is possible to create arrays on any disk on any type of controller in the system, or to move disks from one controller to another.
  • Ease of administration. There is only one software interface to learn: zpool(1M) or mdadm(8). No need to install proprietary vendor tools, or to reboot into BIOSes to manage arrays.
  • Cost. Obviously cheaper since there is no hardware RAID controller to buy.

Consequently, many ZFS and Linux MD RAID users, such as me, look for non-RAID controllers that are simply reliable, fast, cheap, and otherwise come with no bells and whistles. Most motherboards have up to 4 or 6 onboard ports (be sure to always enable AHCI mode in the BIOS as it is the best designed hardware interface that a chip can present to the OS to enable maximum performance), but for more than 4 or 6 disks, there are surprisingly not that many choices of controllers. Over the years, I have spent quite some time on the controllers manufacturers' websites, the LKML, linux-ide and ZFS mailing lists, and have established a list of SATA/SAS controllers that are ideal for ZFS or Linux MD RAID. I also included links to online retailers because some of these controllers are not that easy to find online.

The reason the list contains SAS controllers is because they are just as good as an option as SATA controllers: many of them are as inexpensive as SATA controllers (even though they target the enterprise market), they are fully compatible with SATA 3Gbps and 6Gbps disks, and they support all the usual features: hotplug, queueing, etc. A SAS controller typically present SFF-8087 connectors, also known as internal mini SAS, or even iPASS connectors. Up to 4 SATA drives can be connected to such a connector with an SFF-8087 to 4xSATA forward breakout cable (as opposed to reverse breakout). This type of cable usually sells for $15-30. Here are a few links if you have trouble finding them.

There are really only 4 significant manufacturers of discrete non-RAID SATA/SAS controller chips on the market: LSI, Marvell, JMicron, and Silicon Image. Controller cards from Adaptec, Areca, HighPoint, Intel, Supermicro, Tyan, etc, most often use chips from one of these 4 manufacturers.

Here is my list of non-RAID SATA/SAS controllers, from 16-port to 2-port controllers, with the kernel driver used to support them under Linux, and Solaris. There is also limited information on FreeBSD support. I focused on native PCIe controllers only, with very few PCI-X (actually only 1 very popular: 88SX6081). The MB/s/port number in square brackets indicates the maximum practical throughput that can be expected from each SATA port, assuming concurrent I/O on all ports, given the bottleneck of the host link or bus (PCIe or PCI-X). I assumed for all PCIe controllers that only 60-70% of the maximum theoretical PCIe throughput can be achieved, and for all PCI-X controllers that only 80% of the maximum theoretical PCI-X throughput can be achieved on this bus. These assumptions concur with what I have seen in real world benchmarks assuming a Max_Payload_Size setting of either 128 or 256 bytes for PCIe (a common default value), and a more or less default PCI latency timer setting for PCI-X. As of May 2010, modern disks can easily reach 120-130MB/s of sequential throughput at the beginning of the platter, so avoid controllers with a throughput of less than 150MB/s/port if you want to reduce the possibility of bottlenecks to zero.

32 ports

  • [SAS] 4 x switched Marvell 88SE9485, 6Gbps, PCIe (gen2) x16 [150-175MB/s/port]
    [Update 2011-09-29: Availability: $850 $850. This is a HighPoint HBA combining 4 x 8-port Marvell 88SE9485 with PCIe switching technology: RocketRAID 2782.]
    • Linux/Solaris/FreeBSD support: see Marvell 88SE9485 or 88SE9480 below

24 ports

  • [SAS] 3 x switched Marvell 88SE9485, 6Gbps, PCIe (gen2) x16 [200-233MB/s/port]
    [Update 2011-09-29: Availability: $540 $620. This is a HighPoint HBA combining 3 x 8-port Marvell 88SE9485 with PCIe switching technology: RocketRAID 2760A.]
    • Linux/Solaris/FreeBSD support: see Marvell 88SE9485 or 88SE9480 below

16 ports

  • [SAS] LSI SAS2116, 6Gbps, PCIe (gen2) x8 [150-175MB/s/port]
    Availability: $400 $510. LSI HBA based on this chip: LSISAS9200-16e, LSISAS9201-16i. [Update 2010-10-27: only the model with external ports used to be available but now the one with internal ports is available and less expensive.]
  • [SAS] 2 x switched Marvell 88SE9485, 6Gbps, PCIe (gen2) x16 [300-350MB/s/port]
    [Update 2011-09-29: Availability: $450 $480. This is a HighPoint HBA combining 3 x 8-port Marvell 88SE9485 with PCIe switching technology: RocketRAID 2740 and 2744.]
    • Linux/Solaris/FreeBSD support: see Marvell 88SE9485 or 88SE9480 below

8 ports

  • [SAS] Marvell 88SE9485 or 88SE9480, 6Gbps, PCIe (gen2) x8 [300-350MB/s/port]
    Availability: $280. [Update 2011-07-01: Supermicro HBA based on this chip: AOC-SAS2LP-MV8]. Areca HBA based on the 9480: ARC-1320. HighPoint HBA based on the 9485: RocketRAID 272x. Lots of bandwidth available to each port. However it is currently not supported by Solaris. I would recommend the LSI SAS2008 instead, which is cheaper, better supported, and provides just as much bandwidth.
    • Linux support: mvsas (94xx: 2.6.31+, ARC-1320: 2.6.32+)
    • Solaris support: not supported (see 88SE6480)
    • Mac OS X support: [Update 2014-06-26: the only 88SE9485 or 88SE9480 HBAs supported by Mountain Lion (10.8) and up seem to be HighPoint HBAs]
  • [SAS] LSI SAS2008, 6Gbps, PCIe (gen2) x8 [300-350MB/s/port]
    Availability: $130 $140 $180 $220 $220 $230 $240 $290. [Update 2010-12-21: Intel HBA based on this chip: RS2WC080]. Supermicro HBAs based on this chip: AOC-USAS2-L8i AOC-USAS2-L8e (these are 2 "UIO" cards with the electronic components mounted on the other side of the PCB which may not be mechanically compatible with all chassis). LSI HBAs based on this chip: LSISAS9200-8e LSISAS9210-8i LSISAS9211-8i LSISAS9212-4i4e. Lots of bandwidth per port. Good Linux and Solaris support.
  • [SAS] LSI SAS1068E, 3Gbps, PCIe (gen1) x8 [150-175MB/s/port]
    Availability: $110 $120 $150 $150. Intel HBAs based on this chip: SASUC8I. Supermicro HBAs based on this chip: AOC-USAS-L8i AOC-USASLP-L8i (these are 2 "UIO" cards - see warning above.) LSI HBAs based on this chip: LSISAS3081E-R LSISAS3801E. Can provide 150-175MB/s/port of concurrent I/O, which is good enough for HDDs (but not SSDs). Good Linux and Solaris support. This chip is popular because it has very good Solaris support and was chosen by Sun for their second generation Sun Fire X4540 Server "Thumper". However, beware, this chip does not support drives larger than 2TB.
    • Linux support: mptsas
    • Solaris support: mpt
    • FreeBSD support: mpt (supported at least since 7.3)
  • [SATA] Marvell 88SX6081, 3Gbps, PCI-X 64-bit 133MHz [107MB/s/port]
    Availability: $100. Supermicro HBAs based on this chip: AOC-SAT2-MV8 Based on PCI-X, which is an aging technology being replaced with PCIe. The approximate 107MB/s/port of concurrent I/O it supports is a bottleneck with modern HDDs. However this chip is especially popular because it has very good Solaris support and was chosen by Sun for their first generation Sun Fire X4500 Server "Thumper".
    • Linux support: sata_mv (no suspend support)
    • Solaris support: marvell88sx
    • FreeBSD support: ata (supported at least since 7.0, if the hptrr driver is commented out)
  • [SAS] Marvell 88SE6485 or 88SE6480, 3Gbps, PCIe (gen1) x4 [75-88MB/s/port]
    Availability: $100. Supermicro HBAs based on this chip: AOC-SASLP-MV8. The PCIe x4 link is a bottleneck for 8 drives, restricting the concurrent I/O to 75-88MB/s/port. A better and slightly more expensive alternative is the LSI SAS1068E.

4 ports

  • [SAS] LSI SAS2004, 6Gbps, PCIe (gen2) x4 [300-350MB/s/port]
    Availability: $160. LSI HBA based on this chip: LSISAS9211-4i. Quite expensive; I would recommend buying a (cheaper!) 8-port controller.
  • [SAS] LSI SAS1064E, 3Gbps, PCIe (gen1) x8 [300-350MB/s/port]
    Availability: $120 $130. Intel HBA based on this chip: SASWT4I. [Update 2010-10-27: LSI HBA based on this chip: LSISAS3041E-R.] Also quite expensive; I recommend instead buying a cheaper 8-port controller.
    • Linux support: mptsas
    • Solaris support: mpt
    • FreeBSD support: mpt (supported at least since 7.3)
  • [SAS] Marvell 88SE6445 or 88SE6440, 3Gbps, PCIe (gen1) x4 [150-175MB/s/port]
    Availability: $80. Areca HBA based on the 6440: ARC-1300. Adaptec HBA based on the 6440: ASC-1045/1405. Provides good bandwidth at a decent price.
    • Linux support: mvsas (6445: 2.6.25 or 2.6.31 ?, 6440: 2.6.25+, ARC-1300: 2.6.32+)
    • Solaris support: not supported (see 88SE6480)
  • [SATA] Marvell 88SX7042, 3Gbps, PCIe (gen1) x4 [150-175MB/s/port]
    Availability: $70. Adaptec HBA based on this chip: AAR-1430SA. Rosewill HBA based on this chip: RC-218. This is the only 4-port SATA controller supported by Linux providing acceptable throughput to each port. [2010-05-30 update: I bought one for $50 from Newegg in October 2009. Listed at $70 when I wrote this blog. Currently out of stock and listed at $90. Its popularity is spreading...]
  • [SAS] Marvell 88SE6340, 3Gbps, PCIe (gen1) x1 [38-44MB/s/port]
    Hard to find. Only found references to this chip on Marvell's website. Performance is low anyway (38-44MB/s/port).
    • Linux support: mvsas
    • Solaris support: not supported (see 88SE6480)
  • [SATA] Marvell 88SE6145 or 88SE6141, 3Gbps, PCIe (gen1) x1 [38-44MB/s/port]
    Hard to find. Chip seems to be mostly found on motherboards for onboard SATA. Performance is low anyway (38-44MB/s/port).
    • Linux support: ahci
    • Solaris support: ahci
    • FreeBSD support: ahci

2 ports

  • [SATA] Marvell 88SE9128 or 88SE9125 or 88SE9120, 6Gbps, PCIe (gen2) x1 [150-175MB/s/port]
    Availability: $25 $35. HighPoint HBA based on this chip: Rocket 620. LyCOM HBA based on this chip: PE-115. Koutech HBA based on this chip: PESA230. This is the only 2-port chip on the market with no bottleneck caused by the PCIe link at Max_Payload_Size=128. Pretty surprising that it is being sold for such a low price.
    • Linux support: ahci
    • Solaris support: not supported [Update 2010-09-21: Despite being AHCI-compliant, this series of chips seems unsupported by Solaris according to reader comments, see below.]
    • FreeBSD support: ahci
  • [SATA] Marvell 88SE6121, 3Gbps, PCIe (gen1) x1 [75-88MB/s/port]
    Hard to find. Chip seems to be mostly found on motherboards for onboard SATA.
    • Linux support: ahci
    • Solaris support: ahci
    • FreeBSD support: ahci
  • [SATA] JMicron JMB362 or JMB363 or JMB366, 3Gbps, PCIe (gen1) x1 [75-88MB/s/port]
    Availability: $22.
    • Linux support: ahci
    • Solaris support: ahci
    • FreeBSD support: ahci
  • [SATA] SiI3132, 3Gbps, PCIe (gen1) x1 [75-88MB/s/port]
    Availability: $20. Warning: the overall bottleneck of the PCIe link is 150-175MB/s, or 75-88MB/s/port, but the chip has a 110-120MB/s bottleneck per port. So a single SATA device on a single port cannot fully use the 150-175MB/s by itself, it will be bottlenecked at 110-120MB/s.

Finding cards based on these controller chips can be surprisingly difficult (I have had to zoom on product images on newegg.com to read the inscription on the chip before buying), hence the reason I included some links to online retailers.

For reference, the maximum practical throughputs per port I assumed have been computed with these formulas:

  • For PCIe gen2: 300-350MB/s (60-70% of 500MB/s) * pcie-link-width / number-of-ports
  • For PCIe gen1: 150-175MB/s (60-70% of 250MB/s) * pcie-link-width / number-of-ports
  • For PCI-X 64-bit 133MHz: 853MB/s (80% of 1066MB/s) / number-of-ports

To anyone building ZFS or Linux MD RAID storage servers, I recommend to first make use of all onboard AHCI ports on the motherboard. Then put any extra disks on a discrete controller, and I recommend specifically these ones:

  • For a 2-port controller: Marvell 88SE9128 or 88SE9125 or 88SE9120. I do not primarily recommend it because it is SATA 6Gbps, but because it supports PCIe gen2, which allows the controller to handle an overall throughput of at least 300-350MB/s, or 150-175MB/s/port, with a default PCIe Max_Payload_Size setting of 128 bytes. It is also fully AHCI compliant, in other words robust, well-designed, and virtually compatible with all operating systems; a notable exception is Solaris for which I recommend instead the next best controller: JMicron JMB362 or JMB363 or JMB366. The icing on the cake is that cards using these chips are inexpensive (starting from $22, or $11/port).
  • For an 8-port controller: LSI SAS1068E if you are fine with it only supporting drives up to 2TB. Controllers based on this chip can be found inexpensively (starting from $110, or $13.75/port) and are supported out of the box by many current and older Linux and Solaris versions. In fact this chip is the one that Sun used in their second generation Sun Fire X4540 Server "Thumper". The fact that it can support up to 150-175MB/s/port due to the PCIe bottleneck with concurrent I/O on all ports is sufficient for current disks. However if you need more throughput (eg. are using SSDs), or need to use drives larger than 2TB, then go for its more expensive successor, LSI SAS2008, which supports PCI gen2, which should allow for 300-350MB/s/port before hitting the PCIe bottleneck.
mrb Saturday 15 May 2010 at 03:41 am | | Default
Used tags: , , , ,

131 comments

go

very interesting article, thank you for sharing it.

another update:

warning : AOC-SAT2-MV8 SATA controller (8 ports, PCI-X 64-bit) support hard drive up to 1 TB

go, - 28-06-’10 11:50
vlad

The Adonics HBA AD2SA6GPX1 (and its eSATA variant ADSA6GPX1-2E) is actually based on Marvell 88SE9123 chipset. OpenSolaris ahci driver does not support that chipset yet. (ahci driver is attached but the disks connected to the adapter are not visible to the system)

vlad, - 09-07-’10 13:26
mrb

Thanks vlad, I have updated the blog entry.

mrb, - 10-07-’10 22:36
Mark

Drifting slightly, but on i7 systems, using discrete controller(s) connected to the MCH/IOH PCI Express ports might help avoid the 2GB/s DMI interface becoming a bottleneck. This is only likely to be an issue for those building raid / zfs based around SSDs and producing throughput greater than 1GB/s.

Mark, - 17-08-’10 10:45
GSnyder

First-rate writeup, Marc – thanks for working on this. I wish I’d seen it a week ago before I started slogging through all of these issues on my own. :-)

Unfortunately, OpenSolaris (and I assume, Solaris) doesn’t yet (as of August, 2010) support cards based on the Marvell 88SE912x series, even though they are billed as being AHCI compliant. Several users have confirmed that the OS sees the card but not the attached drives.

Currently I’m using a card based on a Silicon Image 3124, but the performance is notably inferior to the built-in SATA ports on the Intel ICH7 motherboard. There is also an OpenSolaris driver issue that produces frequent hesitations in the data flow, not to mention the fact that OpenSolaris doesn’t support this controller’s NCQ.

I think I’m going to give the SuperMicro ACOC-USAS-L8i a try.

GSnyder, - 23-08-’10 02:20
mrb

Thanks GSnyder. Post updated.

Back to the question of software vs. hardware RAID, Intel acknowledges too that software RAID can be faster:
http://www.theregister.co.uk/2010/09/15/..

mrb, - 21-09-’10 20:32
audi

Great article. Very useful information on what controller cards are available. I found this just after I had done my own bit of researching. LSI now has a 16 port internal card for around $400.

LSI SAS 9201-16i

This card will work nicely in a 20+ hard drive case. I plan on getting this card to create a huge backup server using ZFS and OpenSolaris.

audi, - 18-10-’10 14:05
John

Wondering why these are not included, since they are officially supported by Solaris 10.
LSI SAS 3801 EL-S PCI-E Low Profile SAS
LSI SAS 3801 XL-S PCI-X Low Profile SAS

John, - 26-10-’10 02:23
mrb

audi: great find!

john: I missed the PCIe one (and purposefully ignored the PCI-X model). Also I just found yet another one: LSISAS3081E-R (with internal ports).

I have updated my post. Thanks guys.

mrb, - 27-10-’10 03:58
Instigater

I could get 126MB/s read and 115MB/s write on sil3132 and gigabyte i-ram. Tested with intel iometer on raw storage device.

Instigater, (Email ) (URL) - 02-11-’10 12:02
mrb

Instigater: thanks. I updated my “110MB/s” number to “110-120MB/s” as per your observations.

mrb, - 02-11-’10 12:50
Luzipher

Thanks for a great post ! This saved me a few hundred Euros.
Here is a post about how to fit the Supermicro UIO cards nicely into standard PCIe slots:
(http://blog.agdunn.net/?p=391)
Note that the gen2 version (AOC-USAS2-L8i) has some LEDs indicators on its bracket, but they look easily removable (in fact the LEDs sit on the board itself, it’s just a transparent plastic that routes the light to the bracket).

Luzipher, - 05-11-’10 16:21
Brian

I’m trying to understand how many drives I can connect to an ESATA port on a motherboard that says it is using the JMicron JMB362 chip. I vaguely understand that port multiplier technology is involved, but JMicron’s site says this chip “supports” port multiplier technology and I can’t tell if that’s the same thing as saying it includes/implements such functionality or whether the enclosure I buy needs to have some sort of additional controller. Their site says that this chip “Supports 2-port 3.0Gbps SATA II interface and Supports two independent SATA II channels (separate logic and FIFO)” and this motherboard has two ESATA ports, so I think each port is driven by this chip. I just want to know if I can connect 4 drives to one port, two drives to each port, or only one drive to each port. The hard drive enclosures that require your mobo to support port multiplier technology are a lot cheaper, so I’d prefer to get one of those if it would work. Any ideas?

Brian, (Email ) - 01-12-’10 16:07
mrb

This wording means the motherboard is not using a port multiplier, but that it does support an (optional) port multiplier downstream (eg. in an enclosure to allow sharing this sata link between multiple drives). However you do not need to use a port multiplier.

The chip JMB362 supports 2 sata ports. One of them is your eSATA port, the other one is either internal or unused (the pins of the chip are not wired).

mrb, - 01-12-’10 21:30
domiel

Just wondering if you (or any of your readers) know anything about the intel RS2WC080. I’m guessing it’s equivalent to the LSI SAS2008 but it may be a RAID only device with no JBOD mode (I’ve been bitten by that before – LSI 84016E)

Does anyone know about this card?

domiel, - 19-12-’10 20:43
mrb

Good find domiel. The RS2WC080 is definitely based on the LSI SAS2008 per the specs. It looks like it supports JBOD — that’s my gut feeling based on Intel presenting it as an “entry-level RAID” controller.

mrb, - 21-12-’10 19:34
jloe

Why is it that the cards using the mini-sas (iPASS) connectors
only give you one activity LED per iPASS port?

A 16 drive controller card only has 4 activity LEDs!

Thanks,

Joe

jloe, (Email ) - 22-12-’10 01:28
mrb

jloe: I am frustrated by this as well…

mrb, - 22-12-’10 21:21
swmike

AOC-SASLP-MV8 doesn’t work until (reportedly) 2.6.37, before that it’s highly unstable and won’t even survive an md resync. SMART only makes it worse.

swmike, - 15-01-’11 06:32
289cui

the SIL3132 has poor performance only 27MB/s with both ports used (3rd HDD of the raidz1-tank is on an ICH6R which runs great with 3 other HDDs for another tank, 80 MB/s over LAN which is my limiting factor) running EON/opensoloris b130. luckily I have no bad blocks so no bus reset problems.
under windows xp the SIL3132 does 110MB/s with only 1 HDD connected.

289cui, - 16-01-’11 16:22
mrb

289cui: something is definitely wrong with your setup. A Sil3132 in one of my OpenSolaris 2009.06 boxes achieves 100MB/s with 1 drive, and 145MB/s with 2 drives, confirming the perf numbers in my post.

mrb, - 16-01-’11 18:48
domiel

Has anyone tried using either the SAS2008 or LSI1068E based controllers with the new 3TB WD drives? I emailed LSI but they said that they haven’t tested either of them as yet.

As an aside if anyone is using the new 7K3000’s what do you think of them as there is scant info out there as to how they perform.

domiel, - 17-01-’11 19:42
domiel

Sorry, a correction – that should be 3TB Hitachi drives… which is also the model I referred to as 7K3000

domiel, - 17-01-’11 20:47
Kristleifur Daðason

Thank you! SUPER useful summary of everything.

I can attest that the old PCI-X 88SX6081 Marvell SATA chip in the AOC-SAT2-MV8 works really well under Linux. LSI PCIe cards have worked extremely well for me as well.

(Haven’t had any luck with controllers from Promise so far, not under Linux at least. Do they use their own chips?)

Kristleifur Daðason, - 23-01-’11 12:09
mrb

Kristleifur: Good question… In the early days of SATA, circa 2003, Promise used to design many SATA controller chips and cards. But nowadays they seem to have scaled back this activity. I find very little information about PCIe Promise SATA/SAS cards. It is unclear whether they are well supported by Linux/Solaris/BSD and which chips the card use. Their website is grossly lacking in information.

mrb, - 23-01-’11 19:56
289cui

@mrb
strange and your opensol 2009.06 is even older than EON 0.600 snv_130 (2010-04). sometimes I got 35-37 MB/s with some flukes over 40MB/s.
the hdds are 3x F2 Samsung 1.5 TB, maybe a bad modell ?
I just don’t know what to do, because the the other 3 hdds on the intel-southbridge are rockin.

289cui, - 28-02-’11 05:12
svrocket

mrb –

enjoyed this blog post mucho for building my ZFS NAS/SAN beast. For those that like to live dangerously, dealextreme sells dodgy second tier QA castoff LSI 3041s and 3081s for pretty cheap. But if its DOA, good luck shipping it back to China!

http://www.dealextreme.com/p/lsi-sas3041..

svrocket, - 08-03-’11 23:36
289cui

dmesg is showing me a notice concerning IRQ sharing with my SIL3132, normally this shouldn´t be a problem, but it seems in my case it is.

so my idea was buying a JMB363, which uses ahci as driver in opensolaris, and hoping the ahci driver has no problem with irq-sharing, I tested opensolaris on my workstation with a jmb363 chip onboard, it worked.

so I bought a add-on card PCIe 1x based on JMB363 chip, unfortunataly the card is recognized as “raid-controller” (class-code 10400 not 10601 which is SATA AHCI 1.0) so the ahci driver is not attached.
so I went to the add-on-cards BIOS, hdds are configured as “non-raid” and found no option the select legacy IDE, AHCI or RAID like in the BIOS of workstation-mainboard with the onboard 363.
flashed newer option-rom/bios for the add-on-card, but stil no option and no AHCI.

windows xp and 7 also recognize the jmb363 based add-on-card as raid-class-controller.

but linux (partition magic 5.10) as two devices:
1 IDE (nothing connected)
1 AHCI (my 2 hdds and ahci-driver attached)

which would be nice for me if I wanna use linux (but I don’t) – but bad for people who actually use the card as raid (native by the chip and not via softraid under linux) if there is no switch to avoid this.

so how to achieve this under opensolaris ?
linux seems reading/writing pci(e)-register in “early stages” in linux/drivers/pci/quirks.c (http://lxr.linux.no/#linux v2.6.38/drivers/pci/quirks.c#L1478)

289cui, - 20-03-’11 11:15
svrocket

hey folks I retract my DX link – the $72/$95 LSI cards are counterfeit and have been pulled.

Some Unlucky Buyers posted a about it thread here: http://hardforum.com/showthread.php?t=15..
with lots of hi-res pictures.

svrocket, - 23-03-’11 11:12
Antonio

Hi, by any chance have you tested Dell’s SAS6/iR. Seems to have an LSI1068E chipset. I wonder if it works with FreeNAS (FreeBSD 7 or 8 based).

http://accessories.us.dell.com/sna/produ..

Antonio, - 29-03-’11 11:02
mrb

No, I have not tested Dell’s SAS6/iR.

mrb, - 29-03-’11 23:33
Mar Zah

I am trying to create several multi terabyte arrays and have run into some problems.
I have been running several tests using an assortment of Sans Digital esata JBOD boxes and several RAID cards and have come up with measurements that some people might find useful
I am using 4 Seagate Barracuda Green drives
The tests measure the resync speed of an mdadm RAID10 array with no other processes running on an ASUS P5Q motherboard. I find this is test gives me a useful idea of the real world performance of the array under load
I used the latest bios and drivers for each card
The following numbers are with the drives in a Sans Digital TR4M with its port multiplier. I also used a TR5M and found the numbers identical
Highpoint RocketRAID 2312 resync 85MBps
Cheap SIL3132 card resync 56MBps
Highpoint RocketRAID 622 resync 123MBps
Highpoint RocketRAID 2722 resync 93MBps

The following numbers are with the drives each connected to their own sata cables (no port multiplier)
Using a pair of cheap SIL3132 controller resync 97MBps peaking at 130MBps
Using the P5Q onboard ICH10 controller 270MBps
Using a Supermicro AOC-USAS2-L8i controller 270MBps
Obviously the ICH10 and AOC-USAS2-L8i are limited by the throughput of the drives which I tested at ~130MBps

My question is why is the performance so poor through the port multiplier.

I have been exchanging emails with Highpoint about the very poor results from the RocketRaid 2722. Why would you pay $300 for a card that is outperformed by the free one Sans Digital supplies. Their support department seems clueless and I have to assume the card is basically crap where esata is concerned

Mar Zah, - 14-04-’11 13:06
Michael

Hi, great blog! I have posted links to this blog on several big forums. :o)

A question: I am looking to connect a fast SSD (500MB/sec) as cheap as possible, so I need a SATA 6Gbps card with two slots. Do you know of any such card, supported by OpenSolaris?

Michael, - 24-04-’11 14:35
mrb

Michael: no I don’t know any 2-port card matching your criteria. Your less expensive 6Gbps option for Solaris is the (8-port) LSI SAS2008.

mrb, - 24-04-’11 23:06
Christophe B

Hello, great article. I am rebuilding raids on my box after changing some disks and aligned partitions and have a few questions:

Motherboard is ASUSTeK P7P55D DELUXE (http://www.asus.com/Motherboards/Intel_Socket_1156/P7P55D_Deluxe/#specifications) with 3 SATA controllers: Intel® P55 Express Chipset with 6 ports, JMicron JMB363 with 1 port and JMicron JMB322 with 2 ports.

My setup is 2×1Tb drive as Raid1 for the system (in fact 3 raid1 with different partitions) and 4×2Tb as Raid5 for data.

How should I plug my disks ? All 6 on the integrated Intel chipset ? Or should I use the jmicron to plug the raid1 ? Or spread the raid5 disks across all the controllers ?

Christophe B, - 11-05-’11 06:52
mrb

Put the 6 disks on the P55 chipset’s sata ports. They are all 3Gbit/s & AHCI, and the chipset is connected to the CPU via a 16Gbit/s link (2GB/s), offering enough bandwidth to all 6 disks.

mrb, - 12-05-’11 11:01
rm

New SATA3 chip on the horizon, Asmedia ASM1061:
http://cgi.ebay.com/New-PCI-E-SATA-3-0-SATA-III-Card-6Gb-s-eSata-ASM1061-/150600961598
tried it yet?

rm, (URL) - 16-05-’11 00:28
Christophe B

mrb: Thank you for your answer, I’ve done that and my box is speedlight fast with partition alignment now !

I’ve got a last question: If I want to add two more disks, should I plug them on the two JMicron JMB362 ports or use a spare Promise SATA300 TX4 card on PCI bus ?
I read that PCI bus limits the speed to 266 MB/s so I suppose that is is more interresting to use the JMB362 plugged on the motherboard ?

Christophe B, - 16-05-’11 02:55
mrb

Classic PCI is 32-bit 33MHz, so 133MB/s theoretical peak. Whereas the JMB362 is on a PCIe x1 2.5Gbit/s, so 250MB/s theoretical peak usable (taking into account 10b/8b encoding).

Obviously connecting them to the JMB362 is better. Not mentioning thhat the SATA300 TX4 is an ancient card.

mrb, - 17-05-’11 00:11
Llew

Fantastic post – Helped me to no end!

88SE9125 works with FreeBSD but none of it’s ports (tried PCBSD 8.2 , Freenas 8, GhostBSD 2).

Llew, - 06-06-’11 18:42
Mario

Thanks alot, this was very helpful.

Mario, - 13-06-’11 04:28
Federico

Looks like the 2-port SYBA card with the JM363 has been replaced with http://www.newegg.com/Product/Product.as..

Federico, - 15-06-’11 09:07
Joel

Thanks for the excellent writeup! Helped me locate a 2 port controller card for a new ZFS build.
http://www.newegg.com/Product/Product.as..

Joel, - 17-06-’11 00:26
ldoodle

Hi,

Does anyone know if the Marvell 9480 chip is now supported in Solaris (11 Express). I have seen the Supermicro AOC-SAS2LP-MV8 (http://www.supermicro.com/products/accessories/addon/AOC-SAS2LP-MV8.cfm) which seems absolutely ideal for ZFS seeing as it has no RAID whatsoever.

ldoodle, - 23-06-’11 03:10
ldoodle

Any comments on my previous post. This article is over a year old now so data provided could be well out of date.

ldoodle, - 01-07-’11 06:52
mrb

The AOC-SAS2LP-MV2 is a new card based on the (same old) 88SE9480 controller that was already in my list, shown as not supported by Solaris. (I added the card, thanks).

A quick online search tells me that Solaris still doesn’t seem to support it.

mrb, - 01-07-’11 11:31
ldoodle

Thanks mrb. I thought that would be the case, just wasn’t sure if the 9480 controller was added when the article was written or it is was updated more recently.

ldoodle, - 05-07-’11 03:42
Easy Company

Thanks for keeping this list up to date!

I refer to it from time to time, to assess my upgrade options

Easy Company, - 10-07-’11 21:24
RU_Taurus

Support for various Marvell chipsets in FreeBSD 8.2:
http://www.freebsd.org/cgi/man.cgi?query=mvs&apropos=0&sektion=0&manpath=FreeBSD+8.2-RELEASE&format=html

RU_Taurus, - 27-07-’11 08:58
Sam

Could anyone add a comment on how SAS expanders affect things –

do they introduce a large performance penalty like port multipliers?

Sam, - 22-08-’11 11:08
Ed Swindelles

I’d also like to see some real world numbers about SAS expanders.

Regarding Dell…their H200 uses the LSI 2008 chipset.

Ed Swindelles, (URL) - 22-08-’11 18:54
Andre Reis

If you’re in Europe, StarTech.com sells the (2 port) PEXSATA22I for €29,89 VAT, chipset is JMicron JMB363.

Amazon.co.uk also sells these for ~£21.

Andre Reis, - 03-09-’11 10:39
Ben

Found some experts!

So, I have: – Ubuntu 64bit – P67 MB – mdadm SSD RAID1 array on P67 ports – mdadm 10 disk RAID6 array on P67 and SIL3132 PCI-E port multiplier.

The system is prone to 5 second hangups. It seems to have something to do with the SIL3132 PCI-E card as it’s fine without this card. There may be conflict with my no-name PCI TV tuner devices.

So, my question, can anyone recommend a port multiplier that will work with the Intel P67 SATA ports? I want to use mdadm so I want to be able to address each disk individually. I bought ADDONICS AD5SAPM (SIL chip) but when plugged into the P67 ports I can only can see the first device. I tried the AD5SAPM with a spare JB363 adapter and again could only see the first device.

Any wisdom much appreciated.

thanks
ben

Ben, - 03-09-’11 23:27
mrb

The Sil3132 is not a port multiplier. It is a SATA HBA. On the other hand, the AD5SAPM is a port multiplier.

I would in general recommend avoiding port multipliers. They are supposed to follow specs, and should work everywhere, but in practice there are many bugs, in SATA controllers, and in OS drivers. Some combinations do work, but you are going to spend a lot of time researching that. Your situation seems much easier to solve by finding a SATA HBA that doesn’t cause these “hangups”. Make sure AHCI is enabled for the P67 SATA ports. Try upgrading your kernel/Ubuntu version if you are running a really old one like 8.04. Try removing the tuner card. Try replacing the Sil3132 with a Marvell 88SE9128.

mrb, - 04-09-’11 17:58
Andre Reis

Solaris users beware: I haven’t been able to use the JMB363 with Solaris. The cards I purchased from StarTech.com aren’t set to AHCI, and Solaris sees them as unrecognized RAID controllers.

I’m trying to flash a newer Option-ROM from JMicron, but so far no success.

Andre Reis, - 08-09-’11 05:01
Andre Reis

If you have a PCI card based on Sil3114, it just needs a ROM flash for Solaris to recognize it, see http://kate-ward.blogspot.com/2009/05/i-..

Andre Reis, - 19-09-’11 10:30
Leo

Nice!

Leo, - 22-09-’11 18:48
Martin Gustafsson

Hi,
This is my be the right place to ask for some help regarding controllers for Linux and BSD :)
I got a SuperMicro MBD-X7SPA-H-D525 -O that I use for my home NAS.
It’s time to add some more disks and I’m looking for a controller that I can use for my motherbord. I think I need a PCIe x4 controller even if motherbord have an 16x slot?

Any sugestion on a good controller for the MBD-X7SPA-H motherbord with 8-port or more that I can use for Debian 6.x or FreeNAS 7.x or 8.x.

Thanks Martin

Martin Gustafsson, - 28-09-’11 02:12
mrb

Martin: yes one can plug an x4 card in an x16 slot. I do this myself in some machines. This is called up-plugging by the PCIe standard. Per my list, I would recommend you either an LSI SAS2008 or SAS1068E controller. Note that your motherboard’s mechanical x16 slot is actually only wired as a x4 link, so it will be the I/O throughput bottleneck (2 GB/s in each direction, which is probably far above your needs, so that’s fine).

mrb, - 28-09-’11 19:04
Jim West

I have an HP Proliant DL585 G2 server, and I have an external SATA III RAID 5 array … I have tried 2 different PCIe SATA III controllers, and am only able to support SATA II 3 Gb/sec, since the PCIe slots are version 1.0. Would it be possible to use the PCI-X 64 bit 100 MHz slot with a PCI-X SATA adapter to obtain 6 Gb/s. I have not been able to find a PCI-X 64 bit SATA III controller. I would really like to connect my SATA III RAID 5 array to my DL585 G2 at 6 Gb/s, but not sure how. I would really apprecaite any advice that you might have on this problem.
Thanks,
Jim West

Jim West, - 28-09-’11 19:32
Michael

News:
The IBM M1015 card use the LSI2008 chipset. The m1015 card is cheap and works with Solaris and FreeBSD. Metaluna explains:

Option 1: No Flash Works fine with OpenIndiana, Solaris Works fine with JBOD drives Requires installation of 9240 driver from LSI website May have had a problem with some ESXi installations but now this is fixed?
ANSWER: IT SEEMS TO HAVE BEEN FIXED WITH ESXi

Option 2: Flash to IT mode

Works fine with OpenIndiana, Solaris, ESXi No extra driver installation required. Drivers are built in to OI, Solaris. Possibly better likelihood of working with more OS kernels out-of-the-box Flash procedure is here: http://lime-technology.com/forum/index.p..
Michael, - 29-09-’11 01:59
Michael

Regarding SAS expanders and Solaris and Norco 4224, these two works and are 6GB. Both are using the new generation 6GB LSI Expander chip, so both works with LSI 2008 chipset.

Intel RES2V240

Chenbro AC UEK-23601

Michael, - 29-09-’11 02:01
Martin Gustafsson

Thanks for your feedback mrb. I have just place an order on a new controller for my home NAS. With your help and your posts abow I did order an LSI MegaRAID SAS3081E-R. Hear in Sweden the price one a controller that use the SAS1068E is the same. I just compear the Intel SASUC81 and the LSI SAS30881E-R.
Once agen Thank you.

Martin Gustafsson, - 29-09-’11 03:47
Royce Williams

FreeBSD 8.2-STABLE supports SAS2008 in JBOD mode with the mps(4) driver, as of February 2011. The commit notes are here:

http://lists.freebsd.org/pipermail/freeb..
http://lists.freebsd.org/pipermail/freeb..

If you are running 7.4, 8.1 or 8.2-RELEASE, and you want to run the updated mpt driver as a kernel module, the eRacks guys have a nice howto here:

http://blog.eracks.com/2011/08/lsi-logic..

Ken also noted that LSI would be releasing their own FreeBSD driver, and the rest of the thread after that commit note has some caveats re: disk selection and other items to be aware of.

I can say that mps(4) as a kernel module is working fine for me so far – running 8.1-RELEASE amd64 with a Dell PERC H200, flashed to LSI IT firmware v9.00.00, using ZFS on a pair of WD Caviar Black 2Tib.

Royce Williams, - 29-09-’11 08:26
rektide

highpoint has some new offerings in the pocketraid 27xx.

the low end has the pocketraid 272x, which is a single chip marvell 88SE9485 for ~$180.

higher up models are a multi-chip marvell 88SE9485 PLX bridge solution with a PCIe 16x interface, and 16 & 24 ports. the dual chip offerings are low profile & ~$400 (2740), while the quad 88SE9485 solutions (2760a) are ~$550 . if you want a whole lot of SATA controllers on a single card with a pci bridge, wow, hells yeah, here’s your iops density. tweaktown has a solid review, which covers the “throw more controllers at it” solution used here:
http://www.tweaktown.com/reviews/3553/hi..

major awesome list, thanks for your maintenance. happy hunting throughput warriors.

rektide, (URL) - 29-09-’11 13:11
mrb

Wow these new RocketRAID 27xx are sick. I have always wanted cards to use PCIe switching technology to increase the number of chips & ports… HighPoint is finally the first to do it.

List updated. Title updated to “32 to 2 ports” :) Thanks rektide, and Royce. To Jim: don’t bother with the aging PCI-X tech. You should try a PCIe 2.0 card. It will fall back to 1.0 on your server but that doesn’t matter as the the 3 vs 6 Gbps SATA/SAS link speed is completely independent of the PCIe speed.

mrb, - 29-09-’11 23:26
Jason

Some comments on Newegg suggest that the Intel rebrand of LSI SAS1068E can’t handle drives larger than 2TB. I have not verified this claim personally as I do not have a card yet, but this alone is enough to push me to a LSI SAS2008 based card like the LSI 9211-8i.

Jason, - 05-10-’11 17:05
Rob

Beware Highpoint 27xx has no fBSD support beyond 8.2, and their support@ confirmed no plans for 9.x

Rob, - 26-11-’11 05:17
Tanel

I want to warn for the HighPoint Rocket 620 (Marvell 88SE9128) and the sometimes integrated Marvell 88SE9172 on FreeBSD.

There seems to be an issue somewhere in the layers of ZFS, the AHCI driver or hardware. More about it can be read from:
http://lists.freebsd.org/pipermail/freeb..
http://forums.freebsd.org/showthread.php..

It could of course be related to that specific model of the motherboard, so it would be of great help if someone could share his/her experiences of HighPoint Rocket 620 (Marvell 88SE9128) or Marvell 88SE9172 on FreeBSD.

Tanel, - 12-12-’11 02:25
JimmyZ

Great article, I’ve been looking for a summary like this for years, I still want to know if they support 3TB though.

to go: you said AOC-SAT2-MV8 supports disk up to 1TB, well I have a card using the same Marvell 88SX6081 chip but it can handle 2TB drives, so I guess this is more like a firmware issue.

JimmyZ, - 04-01-’12 07:26
Raman

I have LSI 3801E (1068E chip) and it in fact do not support 3TB drives! I’ve flashed it with latest “IT” firmware (01.33.00.00) and bios (06.36.00.00) and still no luck. Drive reports size as 2,2 Tb…

Are there any options? Perhaps alternate firmware?..

Raman, - 04-01-’12 12:06
Stefano Del Corno

First of all, great page. Lots of usefull details that will help me in the future.. :D

I’d love to share some of my findings with you…

1) OpenIndiana (OpenSolaris) doesn’t handle port multipliers very well. In my experience neither SIL 3132 nor JMicron 363 based HBA work well with SIL port multipliers. I’ve been told that AHCI driver may handle PMs but never had the opportunity to test it.

2) Linux (I did some tests with recent kernels) on the other hand handles PMs very well using both the SIL and the JMicron 363 HBA. If you are performance-wise, avoid PMs, indeed.

3) There is a source of cheap SAS3041 hba in Germany. Look for LSI00168 on ebay and enjoy a decent HBA for 50euro. I’ve already had my pair of boards… The bid will last ‘till end of month, so if you are in Europe, hurry up! :D

Stefano Del Corno, - 07-01-’12 12:17
Ward Vandewege

Quick note – the mvsas driver is unstable/unusable on Debian Squeeze (2.6.32), at least with this card: Supermicro AOC-SASLP-MV8. Avoid. Apparently there are critical bugs in the mvsas driver that are fixed in 2.6.36 or thereabouts.

2012-02-14

Ward Vandewege, - 14-02-’12 14:22
Hiren

Great page –

Thanks for all the information. It is a great summary and interesting to see that there are only a handful of “real” chip vendors (the board makers are abundant).

hiren

Hiren, (URL) - 22-02-’12 14:33
Fizzer

Quote: Antonio
Hi, by any chance have you tested Dell’s SAS6/iR. Seems to have an LSI1068E chipset. I wonder if it works with FreeNAS

Seams to work fine, put in my desktop PC alongside my pci-e graphics card and W7 installed drivers straight off..

Will report back on in in my Ubuntu server, then will see if Solaris Express recognises it.

Thanks for posting this Marc , i was looking at pci stat cards when I came across this. Never thought of it.

Got my Dell SAS6/iR PCIe off fleaBay for £31 + £7 P+P from USA.
Model: http://search.dell.co.uk/1/2/1191-dell-s..

Bargain compared to 8-port PCI-e SATA cards.

Also, first boot in my GigaButt board was a no,no..
Aparently these cards/or SMBus conflicts with Intel motherboard’s memory detection.

Found this site, overclocking forum site that mod them.. LOL Loads of stuff on PERC cards. Hope ya don’t mind the link.

http://www.overclock.net/t/359025/perc-5..

Thanks a bunch.. Mate!

Will report more when I have my ZFS RAIDz pools up, complete n00bie to ZFS, so it may be some time… :D

Fizzer, - 01-03-’12 12:19
rektide

ya’ll ready to let the PCIe v3.0 game start switching in?!

can’t wait till the hba and flash controller become the same chip. and there are four of em. way to go marvell with the scale out answers!

rektide, (URL) - 15-03-’12 19:27
Steve

Thank you for this article. I found it while searching google and now have some ideas for what to buy if/when I decide to build a NAS.

I’m curious though if you happen to have any recommendations for internal drive cages?

Most of my searches bring up external drive cages, or 5.25” drive bay to 3.5” converters.

Ideally, I’m actually looking for something to sandwich in as many 3.5” drives as possible, into as small an area as possible.

Steve, - 22-03-’12 15:32
Gary

Greetings,

An excellent article. Thank you very much. It doesn’t quite answer my particular question.

Do you, or anyone, know of a 4-port SAS/SATA PCIe v2, at least x4, that will support SATA port multipliers based on the SiI3726?

Syba seems to produce a bunch of these 4 port SATARAID’, but can be configured to be plain old SATA controller, that can support such port multipliers. However, they are all 1 lane cards. That works out to be 1.25Gb/port, nominal, and then spread that over 5 drives…ugh. Would be better to stick them on a USB-SATA widget. I figure a x4 card would get me right in the 100MB/s per drive which is in the neighborhood of being tolerable.

All that said, I actually wouldn’t even mind spending the ‘big’ $$$ for a highend raid controller PCIe x8 but I haven’t found any that will support SATA port multipliers.

Any assistance would be greatly appreciated.
Thanks,
—Gary

Gary, - 30-03-’12 00:18
mrb

Steve: I don’t know the market of internal drive bays that well, but I would love to buy this external one for one of my next builds: http://www.plinkusa.net/websata350.htm

5 × 3.5” HDDs in 3 × 5.25” bays, and minimalist design that does not use trays —just slide the drives in!

Gary: in the early years of SATA, port multipliers (PMP) were not very well supported, but these days most modern SATA controllers supports them. For PMP support on Linux, see the PMP column in: https://ata.wiki.kernel.org/articles/s/a/t/SATA_hardware_features_8af2.html

mrb, - 30-03-’12 03:28
Gary

MRB: Thanks for responding. On that and your advice in the article, I elected the HighPoint 2720SGL, since I already had 8087s from my Adaptec card. In any case, as you mention this card is based on the Marvell 88SE9485 and is recognized and by the later linux kernels w/out issue. The mvsas driver loads, sees it and all seems right and true in the world.

I flashed the bios to the latest on HighPoint’s site for the card, v1.4. Note that I tried this in the stock fw of v1.0 w/not change. The issue is that the card/module seems like it’s trying to work, but can’t deal with the port multiplier. Inquire failures and all manner of unhappiness. Interestingly enough, the drive in port 0 of the pmp is seen in the early config menu of the HBA but it can’t even do that much when in linux.

I contacted linux@highpoint-tech.com, which was apparently forwarded to generalsales as that’s who vended me this:

“The RocketRAID 2720SGL is a 6Gb/s SAS RAID controller – it does not support port multiplier devices (PM is limited to certain types of SATA controllers). Our RR64x series, for example, support PM capable enclosures and chassis.”

That flies in the face of the data both from google queries, here, and Newegg’s page, where I purchased the card, here:
http://www.newegg.com/Product/Product.as..

On the “details” tab, note:
BIOS Boot to RAID with point-to-point connectivity or through port multiplier.”

There’s a non RAID BIOS floating around that I found browsing the FreeNAS forums. It’s not clear if this will make a difference as I won’t be able to do anything with it until later tonight.

Any thoughts?
Thanks,
—Gary

Gary, - 10-04-’12 12:23
mrb

Gary: no idea. I have never tried port multipliers in combination with SAS cards.

mrb, - 21-04-’12 01:51
Trygve Sørdal

Port multiplier is SATA only.
Port expander is SAS, with possibility for SATA tunneling protocol .
HP has a 36 port one that is said to be quite good, gives 32 disk capacity.
http://hardforum.com/showthread.php?t=14..

INTEL-SASUC8I (LSI 1068E) works with INTEL AXX6DRV3GEXP (expander with 6-bay hot-swap). This is quite affordable, but installation in a normal case is a bit fiddly
I have personally tested it with 6 × 1TB SATA drives of various manufacture on Ubuntu server.

Be aware that LSI 1068E supports a maximum of 2 TB per disk according to LSI.
http://kb.lsi.com/KnowledgebaseArticle16..

Trygve Sørdal, - 26-04-’12 04:08
joerg

I would like to report my experience with migrating an existing software RAID 5 (7 disks in total) to a new controller. This possibility was for me one of the main points for choosing a software RAID.

The array was created on an Ubuntu 10.04 using mdadm. The controller is a Promise TX8660. I managed to get driver sources for the 2.6 kernel series from promise. Now I wanted to migrate to Ubuntu 12.04 with a 3.2 kernel and I have lost support for this controller.

I thought, that it would be easy to just use another controller, so I bought an Adaptec 6805E. The controller is well supported and integrated without problems when installing Ubuntu 12.04. I installed without the disks containing the old array attached to the new controller.

When attaching the array to the 6805E controller the disks are recognized and the controller BIOS even reports an legacy raid on the disks. However the controller seems to write metadata onto the disks. The disks are available in Ubuntu, but the RAID could not be discovered.

Back to the old system and the promise controller, I found that the partition table was corrupt. I could repair this from the backup tables using gdisk and finally got the array back. The disks were originally partitioned using parted using gpt partition labels and and unformatted primary partition using the entire disk (all are 2TB Samsung).

So the problem seems to be that the adapters these days like to write metadata messing up the info on the disk.

So I now think going away from the Adaptec and the LSI SAS 9211-8i looks promising to me. However I contacted LSI and according to them it also writes metadata. I suspect the same problem might happen.

What are your experiences when migrating an software raid? Is it me or is this a common problem? I thought a controller should leave freshly attached disks alone and just presents them to the OS until explicitly told do do something.

Would the Supermicro AOC-USAS2-L8e be the right choice for me because it is HBA only? However I would like to avoid the UIO problems. I also like to have a controller that will support more then 2TB per disk and 6GB/s SATA.

best regards, Joerg

joerg, - 17-05-’12 18:47
mrb

My experience with migrating software raid is just as easy as (1) move the disks to another machine, (2) boot, and the mdadm array or zfs pool is redetected automatically.

joerg: your experience is the reason the Adaptec 6805E is not in my post. It is not an “ideal” controller as it seems it cannot act as a true JBOD controller. I suggest you replace it with one of the controllers I recommend in my post. They will not forcefully write metadata and corrupt your existing data.

mrb, - 18-05-’12 00:32
joerg

Hi mbr,

Thanks for the answer. And thanks for this blog in the first place. So much useful information here. I wish I would have found it before ordering the controller. I have seen the LSI before and it’s capabilities and driver support looked quite similar to the Adaptec. So it was a 50/50 decision.

Just to complete my observations: It seems that the Adaptec needs to be explicitly configured to present the disks as JBOD. It seems that JBOD is like yet another RAID mode where the controller manages the physical disks and presents them “under his name” to the OS. I did not create such a JBOD volume explicitly and all I do is the controllers default handling. In the OS, I do not see the original disk’s vendor name (SAMSUNG) and model (HD203WI and HD204UI), but I see them being presented as “Adaptec … “ disks. All this is very strange to me, because this might mean that you cannot attach any kind of disk to this controller if you are interested in keeping its data rather than doing a fresh install.

So this is the official warning for all users of software RAID to stay away from Adaptec copntrollers (It is most likely that it is not only the 6805E that behaves this way). I would even expect that a fresh software RAID with these controllers might cause problems if you want to move it to other vendor’s controllers because of the Adaptec metadata.

I already looked more closely at the alternatives:
LSI SAS 9211-8i
Supermicro AOC-USAS2-L8i or L8e
Highpoint Rocket 2720SGL
ARECA ARC-1320-8i

All seem to be SATA3 with 6GB/S, capable of disks > 2TB, and well supported by Linux. Do they all also support presenting the disk’s SMART info to the OS?

Regarding the LSI I contacted their Support asking if it writes metadata to the disks. They were very fast with their reply. I got the following answer:

“The 9211 does write metadata to each of the disks, so if you want to restore files from a software RAID to this, you will need to first create your RAID volume on the 9211, image the existing software RAID volume, and then restore it to the 9211’s volume. Please note, however, that the 9211 cannot do RAID 5. It can only do RAID 0 and RAID 1. If you need RAID 5 support you should look at the 9260-8i.”

This leaves me somehow suspicious, but it could still mean that the controller only writes metadata when using its RAID functions. Which is fine and which I would have expected. I asked LSI now explicitly if it also writes metadata in case the controller RAID functions are not used or if it is flashed with “initiator target” (IT) firmware that does not contain the RAID functions. The answer is pending and I will report what they say.

mbr, If I understand you correctly, the controller leaves the disks alone if I leave it’s RAID functions alone. Have you or anybody here ever moved a software RAID from other vendor’s controllers to the LSI. Or was it only migrating from one LSI model to another one?

I also wonder if the problem might be related to my partitioning. I use gpt labels and primary partitions spanning the entire disk (created with parted). These partitions are used by madam directly without putting a filesystem onto the single disk. Is this a strange way or the usual way of preparing the disks. To me it sounded logical doing it this way because I need a filesystem on the md volume and not on disk level. However a new controller without seeing a filesystem might think it is an empty disk. But this is only guessing about other causes of problems.

I hope it is ok asking all these details here and writing long posts. You might imagine that I want to keep the array as save as possible now and also avoid another useless buy.

best regards, Joerg

joerg, - 18-05-’12 02:33
joerg

I got some answers from Adaptec and from LSI support. Adaptec confirmed that the 6805E is a Hardware RAID controller and cannot work as a simple HBA. Thus it will always write metadata and manage the attached disks.

LSI support said something similar for the SAS 9211-8i. However, they confirmed that I need to put IT (initiator target) firmware onto the SAS 9211-8i. With the IT firmware it acts as a simple HBA and will not write metadata. With the original Firmware it is a hardware RAID that will write metadata.
The procedure to change the firmware is described in
http://kb.lsi.com/KnowledgebaseArticle16..

I guess I have found a way now how to get my array working on the new machine.

best regards, Joerg

joerg, - 18-05-’12 14:39
joerg

To finalize my experience report: I received the LSI SAS 9211-8i. I exchanged the firmware with the IT variant and everything works smoothly now. The HBA boots quite fast and needs just a couple of seconds to recognize all disks. The array assembles without problems. The disks SMART info is available. Performance of the array is good.

Thanks again for this blog. It was very helpful in finally finding the right solution.

best regards, Joerg

joerg, - 26-05-’12 04:07
mrb

joerg, glad you solved your pb. I myself have used the 9211-8i with the IT firmware, and have used software RAID on top of it, and everything worked flawlessly.

mrb, - 30-05-’12 00:52
ccool

Hi,

Excellent Post by the way.

I was wondering if anyone had a positive experience with esata port multiplier and Solaris Express 11? Right now, I have a SIL3132 controller, but because I am unable to make it work, I am looking for another controller that would allow me to use PMP under Solaris Express 11. If anyone had any info on the subjet, I would greatly appreciated.

Also, I have tried Marvell 88SE9128, but as written here it doesn’t work on Solaris. I have also tried my motherboard esata port, but I guess it doesn’t support PMP. I thought my SIL3132 would, but even though it seems to detect, in the logs, the first drive, it will be unable to use it. I am still making tests, but right now, I am a little disappointed. Also, everything works fine on Win7, so the hardware is ok.

Best Regards

ccool, - 30-05-’12 09:16
josh4trunks

Super helpful, thanks! Tossing out my PCIe card to grab a PCIe 2.0 card

josh4trunks, - 07-06-’12 15:29
Peter Bell

Thank you – a very interesting write-up.

I’m currently using the SuperMicro AOC-USAS2-L8i. You comment on the “UIO” layout of this card. I consider this layout to have one great advantage – in a standard tower case, it puts the components (including heatsink) on the upper surface of the card. I have trouble convincing myself that placing the heatsink on the lower surface of the card is ideal.

Before reading this page, I ordered a Rocket 620 to enable me to finally populate all the bays in my 5in3 drive cages – it seems that I have made a good choice there.

Peter Bell, - 22-06-’12 10:06
compudaze

If I put a LSI SAS2008 card into a PCIe 2.0 ×4 slot will I still get 300-350MB/s per port if I only use one of the SAS ports? aka 4 SATA devices?

compudaze, - 24-06-’12 11:16
mrb

compudaze: yes

mrb, - 24-06-’12 15:28
clinnovo

Nice Post Post ! SAS Training in Hyderabad is provided by clinnovo institute that is a pioneer in offering sas course. Data analysis can be done efficiently with the help of

this course and clinnovo makes it easy to operate SAS while you are trying your hands on the same.

clinnovo, (URL) - 08-08-’12 06:05
context

not sure if anyone still reads these. I have tried the ASM1061 card, the net says it works with FreeBSD 9 though i can say NexentaStor (OpenSolaris), it does NOT see drives connected to it during install.

context, - 09-08-’12 23:43
José

Hi,
Here is my experience.
I’m using a Gigabyte motherboard with controller Z68: – 6 SATA ports (2 SATA 3) : fine with OpenIndiana and Solaris 11
I bought a ASUS U3S6 card with a Marvell 9120 controller and 2 ports, the ports are detected as SCSI by cfgadm but disks are cannot start.

So looking for another card, I decided to buy a JMB366 card.

I tried all the day making it work, impossible ….

Although this model seem to be AHCI compliant, and announced here as the best controller, the card is not detected by OpenIndiana.
I have 4 PCI-e slots, and tried each one.
I’m very disappointed, because I spend 80 € to these 2 cards and they don’t work at all :-(

José, - 21-08-’12 08:03
Greg Varga

Have you (or anyone) looked at the LSI SAS2308 based options? Particularly the LSI SAS 9207-4i4e card?

Any experience with running them in FreeBSD?

Greg Varga, (URL) - 27-08-’12 18:51
crabhero

It helps me a lot when I was ready to buy a Adaptec 1430SA for OpenIndiana.
you says it won’t wok under Solaris,so it also can’t be support under OpenIndiana which is based on OpenSolaris.

with this LIST , we can do ZFS in esxi without breaks.

crabhero, (URL) - 30-08-’12 08:29
Kuba Słodzik

@Jose – I managed to get the StarTech JMB366-based card working with Solaris 11. I got updated firmware from http://www.station-drivers.com/page/jmic.. but rather than install the new firmware, I used the flash update program to erase the card’s BIOS. After doing that, you won’t be able to boot from a drive connected to that card, but Solaris 11 will recognize it as an ahci device. It seems to work very well.

Kuba Słodzik, - 31-08-’12 11:18
José

@Kuba,
I’m going to try,
Finally I bought 2 new cards: – one LSI 3041-E with 4 ports (chip LSI SAS1064E) bought on eBay as mentioned by Stefano Del Corno (search LSI00168). It works like a charm.
Best is using IT firmware if you don’t use RAID technology,but IR firmware word too.
Disks are bootable. – one with chip Sil3124 (4 ports), same problem as 2 previous cards, disks are not recognized by cfgadm:
I tried RAID and non RAID firmware found on Silicon website.

For my opinion, LSI is definitively the best choice and the card is cheap :
40€.

José, - 17-09-’12 06:19
CJ Davies

I’ve just bought all the parts for a FreeNAS build, including a HighPoint Rocket 620 (non-RAID) controller, after reading this post. I discovered whilst I was testing/experimenting that I can’t use smartctl to set the TLER to 7 seconds for any disk connected to the 620, but I can to any disk connected to the motherboard’s controller.

This isn’t necessarily an issue as I have 3x disks that need TLER setting on each power cycle & 3x that don’t, so I can connect 2x of the latter to the 620 & the rest to the motherboard.

But can you (or anybody reading these comments) confirm whether the 620 blocking SCT commands affects behaviour when a read error actually occurs? Eg if one of the disks connected to the 620 has a read error & after 7 seconds sends the appropriate signal back to the controller, will the 620 block this error or will it pass it on up to the motherboard & allow ZFS to handle it sensibly?

CJ Davies, (URL) - 14-11-’12 07:59
Shawn

I just wanted to share my experiences with a couple of these cards. I was using Ubuntu Server then switched to Debian Squeeze. I use mdadm for RAID.

I have two AOC-SAT2-MV8 cards in a Supermicro motherboard which I have been using for several years. After a bug was fixed in the driver back in 2008 or so they’ve been fantastic cards. I’ve used WD RE2/3, Hitachi 5K3000, WD Green/AV-GP/Red drives (500GB/2TB/3TB) with no issues.

I recently purchased an Areca 1320 to connect an external enclosure with 8 disks. I had a mix of drives in it so only 4 of them were in the same array and the other 4 may have had a couple in the same array, can’t remember for sure. I had no issues with this setup. Just this past weekend I rearranged all my drives and now I have 5 RE3 and 3 RE2 500GB drives in a RAID6 array on that card. I’ve had 3 drives drop out over the past couple days. The first drive dropped during the initial sync, the second while I was filling up the array with data (this turned out to be a bad drive and was replaced), and the third during another resync. This appears to be the same issue many people have reported with the Supermicro SAS cards with the mvsas driver. I’m running the current stable kernel in Debian Squeeze so I guess I’ll have to look into patching or updating the kernel.

Shawn, - 18-11-’12 15:07
Ceri

HI, I’m trying to get decent performance out of md in linux, and can’t seem to get it.
The number you quote seem way higher that what I see, so I’m wondering if you know something I don’t. I’m hoping you can point me to
I’m using a LSI SAS-2 Fusion controller to an Oracle/Sun J4410 disk array (24 * 2TB), running over two ports (dm) on Oracle Enterprise Linux (RHEL5) and a 2.6.39 kernel. I’m trying md with raid5, raid6, raid10, and can’t get my sequential writes over 100MB/s, yet the reads are 1.5GB/s, so the pipeline is there. I have 128 processors and 128GB or ram, so CPU and memory are not an issue.

Ceri, - 12-12-’12 08:18
Ceri

Additional info: Its the LSI SAS2008 card, which you say is 300-350MB/port, so I’m not sure why vdbench is reporting 1500MB/s on reads over two ports.
Still, the write performance at 80MB/s seems awfully slow…

Ceri, - 12-12-’12 08:26
mrb

Ceri: what does “running over two ports (dm)” mean? What is dm?

mrb, - 13-12-’12 17:53
mrb

I cannot find much information on how the J4410 cage is architectured.

mrb, - 13-12-’12 17:54
Pro Backup

The more expensive Marvell 9485 chip offers an advantage over LSI SAS2008. The mv94xx Linux driver version 4.0.0.1528N is able to detect PUIS (=POIS/ATA6 power up in standby mode) drives. That is not possible with mpt2sas drivers up to version 15.00.00.00.

Pro Backup, (URL) - 03-01-’13 14:56
Kevin M

Good article! There are a couple more advantages of s/w RAID worth noting: 1. Data durability: ZFS does extensive integrity checking, whereas other RAID solutions (h/w or s/w) will often arbitrarily pick between redundant blocks if they don’t match, making the array susceptible to bit rot. 2. Rebuild time: ZFS only rebuilds in-use blocks, whereas h/w RAID will rebuild all blocks, which can increase the risk of another failure during the rebuild.

Kevin M, - 31-01-’13 19:32
Dan Langille

I have an LSI 9211-8i. I haven’t needed to flash it with the IT firmware to get passthrough on disks.

Am I missing something?

Dan Langille, (URL) - 08-02-’13 11:58
SM

I don’t see much here about the 88SE92xx drivers. The Highpoint Rocket 640L or Syba (or whatever other brand they go by) at this link: http://www.newegg.ca/Product/Product.asp... It’s probably a slight bottleneck, but even though, for 4 ports, PCI-E 2.0 ×2 should provide decent bandwidth for adding four spindle disks. Thoughts?

SM, - 13-02-’13 16:33
Fábio Rabelo

I just instaled an Intel RS2WC080 in a OmniOS system with napp-it .
Before the Intel driver instalation ( exactly the way Intel put on read-me ) using “case1” .
After the instalation, system stops the boot process in “hostname :storage ) and never end it, allready 20 minutes waiting !!
Before the instalation, boot ands in just 2 minutes …
No errors or warnings during install
What I did wrong ??

Fábio Rabelo, - 11-04-’13 12:31
anybody

Thanks for your information.
I’m considering to build a ZFS based NAS, your article is very helpful.
Please keep updating this.

anybody, - 28-04-’13 05:48
xmikus01

According to this commit on GitHub, Marvell 88SE9128 should be supported under Illumos: https://github.com/joyent/illumos-joyent/commit/257c04ecb24858f6d68020a41589306f554ea434

xmikus01, - 27-06-’13 15:34
Amit Katyal

Researching info to build a ZFS based NAS. Excellent article, thank you very much!!!

Amit Katyal, - 04-07-’13 12:55
Patrik Dufresne

Very help ful article.

Patrik Dufresne, - 26-08-’13 16:46
kyij

Little searching and found the Rocket 750 – It has 10 mini sas ports to support 40 drives.. each can be 2TB +

kyij, - 27-08-’13 10:15
Artem Ryabov

Marvell 88SE9128 corrupts data on 2TB+ HDD by recording secondary GPT in the center of the HDD on reboot. You can check your drives this command, which checks the GPT signature on the HDD. It is better to check after a reboot.

for sdn in /dev/sd? ; do echo -n $sdn -
sn=$[`blockdev --getsize $sdn`-1]
[ $sn -lt $[1<<32] ] && echo small && continue
sn=$[$sn%(1<<32)]
st=`dd if=$sdn bs=1b count=1 skip=$sn 2>/dev/null| head -c 8`
[ 'EFI PART' = "$st" ] && echo bad && continue; echo ok ; done

Artem Ryabov, (URL) - 02-09-’13 13:39
mrb

Artem: wow, what a bad bug!

mrb, - 02-09-’13 14:29
Artem Ryabov

mrb: yes. it’s very hard to find, becose data corrpted before OS start. By UEFI bios, as I understood. OS don’t have any info about data was changed.

Artem Ryabov, - 02-09-’13 14:44
William

I used 2 LSI SAS1068E without any issues with Debian 7 and ZoL 0.6.1 with 2TB HDDs, no issues at all. I should note that the chip does seem to support 3TB drives with the latest SuperMicro firmware in HBA mode.

I also have a Debian 6 ZoL 0.6.2 (multi-mirror) system with 8 SSDs on a LSI SAS2008, speed is not the best (600MB/s, same SSDs on a SAS2008 on a SuperMicro Mainboard do ~1000MB/s in RAID10) but still very good – again stable without issues.

Both cards are very hard to source in Europe, i grabbed my USAS cards via geizhals.at (German language hardware price search engine, just put AOC-USAS-L8i in it for the 1068E) and via http://www.server-and-more.com (not sure if they sell to EU endusers though).

William, (URL) - 08-10-’13 09:34
gr1n

It looks like VMware has dropped support for Marvell 88SE9128 in ESXi 5.5. But it worked in 5.1…

gr1n, - 15-10-’13 09:14
sarnold

Awesome blog all around and this article in particular. I’m hoping you can help educate me a bit further on something that I’m having trouble finding information on using my usual Wikipedia and googling.

Can SATA drives do multipath? Is it a standard feature of 6Gb SATA? Most drives? Only a handful of drives? Do multipath capable drives have different physical connectors?

Thanks!

sarnold, - 28-11-’13 23:54
Eric

I have a fairly basic question about the “4 × 8-port Marvell 88SE9485,” which supports 32 drives. I’m wondering how do you get 32 drives connected to this controller on those four mini-SAS ports, since I can only find mostly “SAS to 4-port SATA reverse breakout cables.” I’m assuming it will have to use “SAS to 8-port SATA reverse breakout cables” to get 32 SATA drives connected to this controller.

Eric, - 15-01-’14 07:42
Eric

No wait, I think I see what you mean now. It’s 4 SATA drives per mini-SAS port and there are 8 mini-SAS ports. Is that what you mean? Because I just looked at your link to the product and there are only 6x SFF-8087 ports, so does that mean it’s actually only 24 SATA ports, and not 32 SATA ports, since the 2x external ports are SFF-8088 ports? Thanks

Eric, - 15-01-’14 07:47
mrb

Eric: yes, there are 2 SFF-808x ports per 88SE9485. You can connect 4 drives per SFF-808x (regardless if it is 8087 or 8088).

mrb, - 16-01-’14 21:37
Eric

Awesome. Thanks!

Eric, - 17-01-’14 07:55
Jay

Great article. Really helps. Was wondering if you can answer this.
I’m running with MDADM on 6 SSD drives on a SuperMicro server board. I need more drives.

If I install something like LSI SAS2008 can I just begin plugging extra new drives onto that card and rebuild the array ? Will it play well with the onboard raid drives or do they need to ALL be on the card or on the board as opposed to some on each ?

This is a mission critical environment hosting websites.

Your thoughts ?

Jay

Jay, (URL) - 28-04-’14 19:23
Colin

I’m having a bit of trouble with the RocketRAID 2760A. I can see the drives that I plug in with lsblk but I can’t seem to interact with the devices properly. I’m not sure where I’m going wrong yet.

Colin, - 03-07-’14 10:22
Colin

If you’re having trouble with the HighPoint RocketRAID 2760A and you’re using drives greater the 2TB then the issue is likely that you card has RocketRAID 276x Controller BIOS v1.1. I was using 4TB drives. Version 1.2 and later seem to resolve this issue.

Colin, - 09-07-’14 07:40
javier

Nowadays there is something similar priced in 24 ports like rocketraid2760a?

Thank you.

Pd. Moreless price the 9201-16i but it has only 16 ports…

javier, - 15-07-’14 10:00
DMG

Have you looked at the Adaptec Series 7 cards at all? The price point seems really good for 16 ports.

DMG, - 13-08-’14 09:46
(optional field)
(optional field)
Remember personal info?
Small print: All html tags except <b> and <i> will be removed from your comment. You can make links by just typing the url or mail-address.