Samsung 840 Pro timeouts on Areca RAID

[rkagerer]

I have an Areca ARC-1882ix-4G-24, firmware 1.51, with 4x SSD's in RAID0 for the OS/boot disk.

The SSD's used to be 160GB Intel X25-M G2's and never gave me any problems. A few months ago I upgraded to 256GB Samsung 840 Pro SSD's, with a quarter of the space left as spare area.

After months without issues, I've recently started experiencing Time Out Errors on the SSD's in the array. Exactly 5 sec after the time out, the card marks the volume as failed and removes the disk from the array (which results in my system going down). A hard reboot recovers the array (and seems to flush data which was protected in the battery backed up memory of the controller).

Screenshots with further details are below. I did notice that the most recent time this happened, the affected SSD's "POR Recovery Count" SMART stat had gone up by 1, since the last time it happened. Do the Samsung SSD's go into some kind of "deep recovery" state and do I need to enable some kind of TLER-like feature on them?

Has anyone else had this problem?

Note I did find someone with similar issues in this thread, http://hardforum.com/showthread.php?t=1476948, and mention of disabling Spread Spectrum Clocking. It is disabled for my motherboard, but I'm not sure how to disable that for the SSD's (need to use Samsung Magician?)

When I presented this info to Areca, they advised me to try changing the "Disk Write Cache Mode" from "Auto" to "Enabled". But I'm not sure if the Samsung's have supercaps and I'm concerned about potential data loss from enabling disk write cache. (e.g. In the event of a sudden power loss / system reset I don't want data queued in the cache - which the controller thinks is written - to be lost. Especially since my "fix" to this issue is a hard reset).

Note the Samsung SSD's were recommended to me by other users for use with this card: http://www.xtremesystems.org/forums/...-on-Areca-RAID

This just in: I also seem to be getting occasional, random single- and double-beeps out of the controller with no event logged.
Edit: Apparently Adaptec is tracking an issue with these drives, not sure if it's related: http://hardforum.com/showpost.php?p=...61&postcount=8

Event Log:


Card Settings:


Drive Info for Slot 1:


SMART Stats for Slot 1:


Issue has been cross-posted at HardForum as well: http://hardforum.com/showthread.php?t=1765754

[Computurd]

I'm not sure if this is relevant, but the fact that these Arecas feature the ROC is probably pretty relevant. I haven't read the thread, but im sure this is a good place to read.

http://forums.servethehome.com/raid-...ompatible.html

EDIT: did a bit of reading on it now, and the conspiracy theory makes for an interesting scandal, but not one that i think is intentional as the OP claims. LSI is a huge OEM, I very highly doubt they would risk that, or customers, over shady business like that.

[Highendtoys]

I can help you out in a few days. My card is in a computer that's in transit. When it gets here I'm hooking up 16 840 Pro drives.

From what you said though, it sounds like a cable could be bad. These cables are fragile and any kink will screw them up. I have a number of cables around that have 1 bad end of them.

[ChrisRoad]

We're using eight Samsung 840pro 512GB drives on a Areca 1880 (with 4GB cache module), and well, so far we did not see any timeouts, but the performance is abmysal.
We have configured a RAID6 array over these 8 drives and created a 2.4TB volume (resulting in ~400GB used space on each of the disks, leaving enought overprovisioning space)

It's possible that the very bad performance (latency in the three digits regions and low throughput) comes from the fact that the disk drive cache is disabled, but so far I did not have the chance to forcefully enable the drive cache withing the Arece controller options, as this requires a server reboot that we cannot afford at the moment.

Btw. the Samsung 840 pro drives do also not work on the onboard SCU controller of modern E5 Xeon systems. (C600/602 Chipset)
We did some extensive tests with 12 different 840pro drives in 5 different Xeon Servers (3x Intel, 1x SuperMicro, 1x IBM) and when using them on the SCU (in SATA mode, PassThough or in RAID will not matter) they will just drop out and cause a bus reset after 5-60 seconds.

So I would recommend to stay away from these drives when it comes to some sort of server usage, maybe except for using them as boot device on the onboard SATA/AHCI controller.

[tiro_uspsss]

We're using eight Samsung 840pro 512GB drives on a Areca 1880 (with 4GB cache module), and well, so far we did not see any timeouts, but the performance is abmysal.
We have configured a RAID6 array over these 8 drives and created a 2.4TB volume (resulting in ~400GB used space on each of the disks, leaving enought overprovisioning space)

It's possible that the very bad performance (latency in the three digits regions and low throughput) comes from the fact that the disk drive cache is disabled, but so far I did not have the chance to forcefully enable the drive cache withing the Arece controller options, as this requires a server reboot that we cannot afford at the moment.

Btw. the Samsung 840 pro drives do also not work on the onboard SCU controller of modern E5 Xeon systems. (C600/602 Chipset)
We did some extensive tests with 12 different 840pro drives in 5 different Xeon Servers (3x Intel, 1x SuperMicro, 1x IBM) and when using them on the SCU (in SATA mode, PassThough or in RAID will not matter) they will just drop out and cause a bus reset after 5-60 seconds.

So I would recommend to stay away from these drives when it comes to some sort of server usage, maybe except for using them as boot device on the onboard SATA/AHCI controller.

personally never liked the sammies the..TLC(?) nand flash they have, has significantly less lifespan in terms of write (to each cell) endurance

I'm not sure if this is relevant, but the fact that these Arecas feature the ROC is probably pretty relevant. I haven't read the thread, but im sure this is a good place to read.

http://forums.servethehome.com/raid-...ompatible.html

EDIT: did a bit of reading on it now, and the conspiracy theory makes for an interesting scandal, but not one that i think is intentional as the OP claims. LSI is a huge OEM, I very highly doubt they would risk that, or customers, over shady business like that.

mmmm... it oh so reminds me of the ULi southbridge debacle of the ATi CFX3200 chipset... I still blame nvidia.. screw you nvidia

[rkagerer]

We're using eight Samsung 840pro 512GB drives on a Areca 1880 (with 4GB cache module), and well, so far we did not see any timeouts, but the performance is abmysal.

We did some extensive tests with 12 different 840pro drives in 5 different Xeon Servers (3x Intel, 1x SuperMicro, 1x IBM) and when using them on the SCU (in SATA mode, PassThough or in RAID will not matter) they will just drop out and cause a bus reset after 5-60 seconds.

That's brutal. I went with the 840Pro's based on recommendations on this forum (http://www.xtremesystems.org/forums/...t=#post5171569)

Performance has been acceptable, though not as high as I'd hoped. But the issues with timeouts / disconnects I've been having are annoying. I haven't conclusively traced it to the drives, but I never had any problems at all with the Intel X-25M SSD's that these drives replaced. Because of that, in hindsight, I wouldn't have made the upgrade. Should have waited for something like Intel DC S3700.

[ChrisRoad]

On the Areca we replaced 8x 160GB Intel 320 Series SSDs, and they ran perfectly for two years. (did I mention that these drives do still have 99% wear-out, despite the fact that they were used in a RAID6 without trim in a hispeed datastore for ESXi based virtual machines)

We would have used Intel 320series/600GB oder 520series/480 drives as we've used them (in masses) before and they run very well in any of our configuration.....but Intel somehow refused to ship any these drives for the last couple months...

[davidm71]

Wonder if a solution was figured out since the last post?

[josh1980]

I'm somehow thinking "no".

If I remember correctly some of the Samsung drives don't work with some SATA controllers(mainly Marvell). Apparently the fix is in Samsung's court for those cases and Samsung is saying they aren't going to fix it because it's not their problem. Just check out the 1-egg warnings on Newegg. My guess is Samsung has fubared something and the solution can't be resolved with a firmware fix.

OCZ is in the penalty box right now for trying to undercut everyone in the price arena at all costs. Anyone noticing Samsung is trying to do the same? Anyone notice a correlation here? Enough said...

Meanwhile, i'll add that my 8 Intel SSDs are all coming up on 1-3 years of usage, all are doing great with no problems and no RMAs...

[davidm71]

I had something similar happen with a two drive raid 0 array on my X58 ich10r controller with Sandisk Ultra Plus drives. Only happened once during first run but has me spooked. Interesting to note that Sandisk uses Marvel for its ssd firmware..

[rkagerer]

I know it's been a while but I'm just pinging back here to report that I still do get intermittent timeouts. Once every few months or so. Two in the last few days. Looking to replace them with the Intel S3700 or S3500.

[Computurd]

There are a few options out there that are great for light-enterprise duty. the M500DC in particular is pretty good, better than the DC S3500 in most workloads