Server Class SSD SAS Disks

  • Has anyone installed or even evaluated SSD SAS disks that are rated as server class for non-SAN direct attached storage. If you have, what brand, model, and vendor did you use? What RAID configuration did you use? Did they work the way you hoped?

    I am looking for SSD with a SAS interface that I can use in places where SAN is not needed, but where high performance for random read and random write would be important.

    I know that I could use a solution like the FusionIO disks, but I am looking for something has a lower price and performance point that could just replace a normal SAS disk.

    Most of the SSDs that I have seen use a SATA interface and seem to be designed for workstation applications where the write load wear is not a big issue.

  • Giving this a bump.

  • I have looked and done some research, but haven't gotten anyting to try yet.

    You might look at what HP offers: http://h20000.www2.hp.com/bc/docs/support/SupportManual/c01580706/c01580706.pdf

    It looks like they are going to have some good options early this year. (I was mainly looking at their option because they would go right into existing hotplug SFF/LFF slots on the HP servers.)

  • I was looking at the following from Hitachi Global Storage Technologies, but I don't think they are widely available yet. Seems to be the general problem for enterprise SSDs; they're not quite on the market yet, but may become more widely available this year.

    Ultrastar SSD400S 400GB, 200GB & 100GB Enterprise SSDs

    http://www.hitachigst.com/solid-state-drives/ultrastar-ssd400s

    The announced specs look very nice, although I would really like to see the numbers for random 64K reads and random 64K writes that are most important for SQL Server. The endurance specs look very good; that's going to be very important to convince people to trust data to these devices. The number I'm most interested in is price, and I can't find anything on that.

    Interface:SAS 6Gb/s

    Read max MB/s sequential 64K:535

    Write max MB/s sequential 64K:500

    Read max IOPS, random 4K:46,000

    Write max IOPS, random 4K:13,000

    Endurance, max PB, random write:35 for 400 GB drive

    Also, Western Digital just acquired Hitachi Global Storage Technologies, so it's hard to say what impact that might have on these products.

    http://www.hitachigst.com/press-room/2011/western-digital-to-acquire-hitachi-global-storage-technologies

  • You can see more specs on the new HP drives on page 11 of this: http://h20000.www2.hp.com/bc/docs/support/SupportManual/c01071496/c01071496.pdf

    The sequential numbers don't look as good, but the random numbers are fairly similar.

  • Yes, I've seen those numbers.

    What I haven't seen is pricing. If I could make the case that a pair of SSD drives mirrored could give me the same performance as 10 SAS disks in a RAID 10 array, then it's just a price/performance comparison as long as the SSD is large enough to hold the database.

  • I read a good article on this at TomsHardware.com a few months back. I just looked at their homepage and they have a link devoted to SSD storage. Sorry I don't have the link to the article I read, but the main page is http://www.tomshardware.com/

    K

  • My understanding is that for most current SQL applications, SSD is used for the TEMP DB (for performance). If the drive dies, swap another one in and forget about it.

    This is also not available just yet, and there are no prices listed, but Seagate is getting serious with SSD. Here is a quote from it:

    Seagate claims the Pulsar XT.2 can sustain up to 10 full drive writes per day every day over its three-year warrantied lifespan, which amounts to 15 petabytes worth of data written to the drive.

    That may not be good enough just yet, but at any rate, we're getting there.

    The greatest enemy of knowledge is not ignorance, it is the illusion of knowledge. - Stephen Hawking

  • mtillman-921105 (3/16/2011)


    My understanding is that for most current SQL applications, SSD is used for the TEMP DB (for performance). If the drive dies, swap another one in and forget about it.

    I wouldn't go with anything less than two SSD units setup in a RAID1 configuration. While it is true there should be very little if any data loss, do you really want your production SQL Server to go down, and have to swap hardware, partition it, format it, setup the directory structure, setup the permissions, and then restart your SQL Server instances that will have to go through the recovery/rollback process on all of your databases? And if you didn't have a spare SSD on hand you would have to wait hours or possibly days for your service vendor to arrive with a replacement before you could start getting things going again.

  • UMG Developer (3/16/2011)


    mtillman-921105 (3/16/2011)


    My understanding is that for most current SQL applications, SSD is used for the TEMP DB (for performance). If the drive dies, swap another one in and forget about it.

    I wouldn't go with anything less than two SSD units setup in a RAID1 configuration. While it is true there should be very little if any data loss, do you really want your production SQL Server to go down, and have to swap hardware, partition it, format it, setup the directory structure, setup the permissions, and then restart your SQL Server instances that will have to go through the recovery/rollback process on all of your databases? And if you didn't have a spare SSD on hand you would have to wait hours or possibly days for your service vendor to arrive with a replacement before you could start getting things going again.

    The rumors of SSD failure rates remind me of a system where we started debating on using RAID 11 out of self defense. Yes, a mirrored mirror.


    - Craig Farrell

    Never stop learning, even if it hurts. Ego bruises are practically mandatory as you learn unless you've never risked enough to make a mistake.

    For better assistance in answering your questions[/url] | Forum Netiquette
    For index/tuning help, follow these directions.[/url] |Tally Tables[/url]

    Twitter: @AnyWayDBA

  • mtillman-921105 (3/16/2011)


    My understanding is that for most current SQL applications, SSD is used for the TEMP DB (for performance). If the drive dies, swap another one in and forget about it.

    This is also not available just yet, and there are no prices listed, but Seagate is getting serious with SSD. Here is a quote from it:

    Seagate claims the Pulsar XT.2 can sustain up to 10 full drive writes per day every day over its three-year warrantied lifespan, which amounts to 15 petabytes worth of data written to the drive.

    That may not be good enough just yet, but at any rate, we're getting there.

    The drives may be OK, but whoever wrote the article didn't know what they are talking about:

    "The Pulsar XT.2 is 50% faster than its predecessor and boasts...an average seek response time of about 10 milliseconds." 10 milliseconds would make it slower than a rotating disk.

    Also, in one place it says they have a 5 year warrenty and in another place it says they have a 3 year warrenty.

  • Michael Valentine Jones (3/16/2011)


    mtillman-921105 (3/16/2011)


    My understanding is that for most current SQL applications, SSD is used for the TEMP DB (for performance). If the drive dies, swap another one in and forget about it.

    This is also not available just yet, and there are no prices listed, but Seagate is getting serious with SSD. Here is a quote from it:

    Seagate claims the Pulsar XT.2 can sustain up to 10 full drive writes per day every day over its three-year warrantied lifespan, which amounts to 15 petabytes worth of data written to the drive.

    That may not be good enough just yet, but at any rate, we're getting there.

    The drives may be OK, but whoever wrote the article didn't know what they are talking about:

    "The Pulsar XT.2 is 50% faster than its predecessor and boasts...an average seek response time of about 10 milliseconds." 10 milliseconds would make it slower than a rotating disk.

    Also, in one place it says they have a 5 year warrenty and in another place it says they have a 3 year warrenty.

    I'm not sure about the seek time spec, if it's a typo or what, but the 3 year / 5 year warranty applies to two different drives. I was confused about that too until I went to Seagate's site and looked at the data sheet for each SSD. The Pulsar .2 comes with a 3 year warranty, but the Pulsar XT.2 comes with a 5 year warranty.

    The greatest enemy of knowledge is not ignorance, it is the illusion of knowledge. - Stephen Hawking

  • UMG Developer (3/16/2011)


    mtillman-921105 (3/16/2011)


    My understanding is that for most current SQL applications, SSD is used for the TEMP DB (for performance). If the drive dies, swap another one in and forget about it.

    I wouldn't go with anything less than two SSD units setup in a RAID1 configuration. While it is true there should be very little if any data loss, do you really want your production SQL Server to go down, and have to swap hardware, partition it, format it, setup the directory structure, setup the permissions, and then restart your SQL Server instances that will have to go through the recovery/rollback process on all of your databases? And if you didn't have a spare SSD on hand you would have to wait hours or possibly days for your service vendor to arrive with a replacement before you could start getting things going again.

    I would agree, although I'm no expert, that SSD needs to improve before it's enterprize ready. But if performance is of utmost importance, SSD is beginning to be a viable alternative.

    Sure, if you have to rely on a service provider to replace a drive, then that's an issue. On the other hand, if you have your own in-house hardware tech, physically replacing a drive shouldn't be an ordeal. (For what it's worth, they claim to be hot-swappable too.) Also, having a spare on hand, maybe just a regular drive, would be advisable too in that scenario.

    The greatest enemy of knowledge is not ignorance, it is the illusion of knowledge. - Stephen Hawking

  • I run fusionIO SSDs - pcie cards. Stripe them together, mirror the databases.

    Absolutely amazing performance.

  • RichB (3/21/2011)


    I run fusionIO SSDs - pcie cards. Stripe them together, mirror the databases.

    Absolutely amazing performance.

    What size/model are you using? How much did they cost?

Viewing 15 posts - 1 through 15 (of 16 total)

You must be logged in to reply to this topic. Login to reply