M1015 or 9240-8i & Samsung Pro 850 Issue

[Tom D]

G'day everyone, all the best for the New Year. I've run into an issue that has me puzzled with my rather antiquated Supermicro X7DWA-N-based workstation using an IBM M1015 or an LSI 9240-8i with Samsung 850 Pro SSD's. I apologise if this issue has been brought up before, but I couldn't find anything about it using the search function.

I originally built this machine back in 2008 with a couple of 3.4 GHz quad core Xeons and 32 GB of RAM. At the time, I simply transferred my LSI dual-channel U320 SCSI card and hard disk RAID-0 array from my previous PC and have been running that for a while up until now.

One of the disks in the array has just cashed its chips in, so I figured it was time to boost the performance a little and look at SSD substitutes. I picked up an IBM M1015 dirt-cheap from a friend and I ended up buying four 128GB Samsung 850 Pro SSD's as the 10 year limited warranty printed on the boxes sounded good (although I don't know how that'll work out in practice), and the price here worked out to less than a buck per Gig. I'll only be using the operating system and programs on this RAID array as I keep my Ghost image, data and documents on two separate SATA mechanical drives, so I didn't think there was much point in opting for the larger capacity SSD's.

I thought I would run some tests initially just to suss out any potential problems, so I started with a single SSD to get a baseline idea of how they would perform. I connected this SSD to the motherboard SATA port running in AHCI mode. I wasn't expecting much from the SATA II ports thinking this drive would saturate the port and came up with the following result:

I then hooked up all four SSD's in RAID-0 on the SATA ports and came up with following result:

The next test was a single SSD as a JBOD drive on the IBM M1015 card (it has the latest firmware and O.S. has latest driver), as you can see the Seq Q32T1 and Seq readings increased over the single SSD on the SATA port.

The next test was four SSD's in RAID-0 on the IBM card, the Seq Q32T1 increasing over the RAID-0 setup on the SATA ports, but dropping below the 4K Q32 Random Read and Write test of the SATA II ports, the IOPS being significantly lower than that of the SATA II test.

I didn't have another card to try and all I could borrow was a new LSI 9240-8i from a friend and ran the tests again. There was no improvement at all, the 9240-8i just mirrored my previous M1015 tests. After double checking everything, it seemed I had no other option but to go out and buy and try a different card, which would take time, so I decided to re-flash my IBM M1015 to a 9211-8i with the latest ir firmware in the interim to see if that would make any difference.

The 9211-8i ir firmware certainly improved the results as you can see in the following test. I'm not sure what's going on here, I really don't know what I should expect to obtain.

I was looking at buying an LSI 9266-8i, for example. Or does anyone have a better recommendation for a card? My board has 2nd generation PCI-e x16 slots, so going to a more recent Gen. 3 PCI-e card may only offer future upgradeability with no real advantage on this machine.

Yeah, I know this is an old rig (like its owner-I'm in my sixties), but it's been rock-stable for years, often doing data logging for days. I'm pretty reluctant to build a new Supermicro Xeon-based PC at present as I would have to import most of the parts (read expensive) and I use this machine with a GPIB bus for my test instruments. A new PCI-e GPIB card for a newer generation motherboard is just over a grand - I'm not too keen on laying out that amount for a card. I still use an old EPROM programmer on the parallel port and have other devices that use the serial COM ports, it's unfortunate that these are now considered to be obsolete and are no longer included on the newer stuff. Any insight would be much appreciated.

Tom

 

[Wondersausage]

The Samsung 850 Pro has a 512MB volatile RAM cache and so LSI wisely disables the drive cache, affecting performance but keeping your data reasonably safe. The Samsungs have a habit of breaking a RAID array on dirty shutdown due to the cache (and that's not solved by a controller BBU because the drive itself needs to retain power, so use a UPS). If you're going to run SSD RAID, it's best to use either a series with a non-caching controller (e.g. Sandforce) or a supercap backed up drive (e.g. Intel 7xx/S3xxx).

If you upgraded the system board you wouldn't need a RAID card since the modern onboard Intel RAID is faster for SSDs than a MegaRAID (Intel supports TRIM pass through). No need for Xeons unless you need more than 8 cores.

There are plenty of USB to Parallel, USB to Serial, and USB to GPIB devices available cheaply. IEEE488 is only 1MBps, no need for PCIe

BTW warranty years are relatively meaningless on SSDs because the industry has subtly changed to warranty by TBW (terabytes written) or years, whichever comes first. The 850 Pro 128GB warranty is 150 TBW, or 82 GB per day.

 

[Tom D]

Thank you for your response, Wondersausage. I do use an Uninterruptible Power Supply on this system. Nearly every time my younger son fires up the MIG welder to work on his Mach 1 Mustang, it invariably trips the circuit breakers at the fuse box - this always seems to happen when I'm using my computer. I could be wrong, but I don't think my machine draws more current than the MIG.

Unfortunately, upgrading the system board for me means upgrading just about everything - motherboard, CPU's, heatsinks-fans, RAM and case. I was somewhat puzzled by the apparent difference in performance converting from a 9240-8i to a 9211-8i by re-flashing the card. I was kinda hoping there was a card someone may have had experience with that was specifically aimed at SSD usage.

My own experiences with USB to Serial adapters have been less than stellar, I've tried five or six different inexpensive units and have had crashes, system freezes or lock-ups. I finally got one that seemed to work with my hand-held scope, but it didn't want to know my bench-top multimeter which is the same brand as the scope.

My current NI GPIB card will only plug into a standard PCI or Low Voltage PCI slot, so speed is not really the issue here. Looking at Supermicro's X10xxx series boards, I don't believe any of them have a PCI slot, everything seems to be PCI-e. National Instruments want something like 700 odd bucks for their GPIB-USB adapter. I know there are cheap - shall we say - "genuine" NI GPIB USB adapters on Ebay everyday of the week from China, but I'm not game enough to buy one of these, the sellers probably get them from the same supplier who supplies their Hello Kitty paraphernalia. Apparently the latest driver checks the serial number of the adapter and reports that it's non-genuine - I don't know if this impacts the functionality.

A little while back, Ebay was flooded with Agilent 82357B GPIB-USB fakes or clones. A friend of mine bought one late last year and although he never got around to performing in-depth testing on the thing, he reckoned that it worked reasonably well... For about three weeks.

As it is at present with the four SSD's on the re-flashed 1015, the machine seems to be pretty stable with no issues so far, so I think this might have to do for the time being.

Tom

 

[sol45]

Tom, can you tell what card you choose and how this working for you? I have same MB and order LSI 9361-8I yesterday(was on sale) going to receive 02/16 i hope this going to work.

 

[Tom D]

Hi sol45,

At present I'm still using the IBM 1015 that I cross-flashed to a 9211-8i. I've just received an LSI 9266-8i from a friend, but I haven't had a chance to fit it and experiment with it as yet, I've been pretty busy assembling my son's BOSS 351 engine for his Mustang. As soon as I get some free time, I'll install the card and post my results here.

Tom