System Performance - SCSI vs SAS

bbrownbbrown Member Posts: 3,268
edited 2008-05-25 in SQL Performance
Has anyone done any comparisons in the performance difference between SCSI and SAS drive systems?

We are running two upgrades (of the same 160 GB SQL database) on two different servers. One server has SCSI drives the other has SAS drives. The SCSI machine has a lot more drives.

The SAS machine is running at close to 3 time as fast in completing the process.

I'm starting to focus on the drive difference being the reason.

Any thoughts?
There are no bugs - only undocumented features.

Comments

  • Slawek_GuzekSlawek_Guzek Member Posts: 1,690
    Hi,

    SAS = Serial Attached SCSI. The difference is only in physical transmission medium, all features giving SCSI performance advantages are present in both standards.

    The difference must be due to something else. If you have many more SCSI disks, and all are on single SCSI bus then you may experience congestion on SCSI bus and performance slowdown, especially with sequential access and bigger transfers.

    You may download SQLIO and run the test yourself, then you'll see how you hardware performs.

    http://www.microsoft.com/downloads/deta ... laylang=en

    Regards,
    Slawek
    Slawek Guzek
    Dynamics NAV, MS SQL Server, Wherescape RED;
    PRINCE2 Practitioner - License GR657010572SG
    GDPR Certified Data Protection Officer - PECB License DPCDPO1025070-2018-03
  • Marije_BrummelMarije_Brummel Member, Moderators Design Patterns Posts: 4,262
    Since Navision is an I/O eater SAS outperforms SCSI. My first experience with SAS was 1 year ago. A HP570 with a capacity of 18!! DAS SAS disks.

    Wow.

    Since then I have experienced the difference over a dozen time. It is so much better.

    For our own performance lab, we have bought a HP 370 with a max. capacity of 16 SAS disks on DAS. We have 2 RAID controlers, one P400 and one P800. Customers and partners in NL and BE can use our performance lab to test their systems on affordable hardware. Our server costs less than 10k. (euro).
  • bbrownbbrown Member Posts: 3,268
    SAS definitely outperforms SCSI in this situation. The SAS system finished the task (upgrade step 1) in 6 hours that the SCSI system took around 15 hours to complete.

    The SCSI system (current production server) has 28 drives across the 2 RAID 10 arrays that support the secondardy data file. The SAS system (test) has only 4 drives in a RAID 10 array. Both systems have separate arrays for transaction logs. The SCSI system has 24 GB memory and the SAS system has 8 GB.

    While the SAS system provide better performance under the sequential nature of upgrade processing, I feel that the SCSI system is stil the better choice for production support of this system. A produstion system tend to involve a lot more random i/o and the multiple drives will support that better. Also this system has been supporting the previous database (3.60 SQL with 5.0 client) for some time. If the SAS sytem had more drives I might be tempted to switch.

    I've always understood SAS to have higher perfromance potential but this was the first time I've had a direct side-by-side comparison. This just continues to reinforce theory #1 - There will never be a "fast enough" server.

    We have another customer that is in the process of a hardware upgrade. They are replacing a 14 drive SCSI array with a 45 drive SAS array. We'll be doing a NAV upgrade this summer. Should be interesting.
    There are no bugs - only undocumented features.
  • Slawek_GuzekSlawek_Guzek Member Posts: 1,690
    Hi,

    IMHO you cannot easily judge SAS vs. SCSI only by results in processing time.

    28 disk = 2 x 14 disk array, minimum 2 SCSI buses. If you have only 2 SCSI buses then 14 disk/bus very likely causes congestion. On the other hand SAS is a PTP interface - all disks can transfer simultaneously (if controller is smart enough)

    Apart from SAS and SCSI differences there might be quite a few other differences: in physical disk parameters (disk rotation speed, disk seek times, disk internal cache), controller settings (amount of cache, cache mode, RAID config details), OS config details etc.


    Regards,
    Slawek
    Slawek Guzek
    Dynamics NAV, MS SQL Server, Wherescape RED;
    PRINCE2 Practitioner - License GR657010572SG
    GDPR Certified Data Protection Officer - PECB License DPCDPO1025070-2018-03
  • bbrownbbrown Member Posts: 3,268
    The systems are similar in most areas. In those that differ the SCSI system has the "larger" configuration.

    They are both 64 bit OS and SQL

    The SCSI system has 4 quad core processors (2.67 MHZ) while the SAS system has 2 quad core (1.87)

    All controllers are set to 100% Read cache

    The SCSI system is using 72 GB 15K 3.5 inch drives. The SAS system is using 146 GB 10K 2.5 inch drives. The SCSI drives has somewhat better specs.

    I've spent a few hours collecting performance data from the two systems. The only noticable difference was the rate at which data was being transferred to and from the drive system. On average the numbers from the SAS system were better then double that of the SCSI system. Also the numbers from the SAS system were fairly steady while the SCSI traffic was somwhat bursty.

    BTW - The SCSI drives are across 4 channels. And yes channel congestion may be an issue. It's quite possible that the 28 drives cannot reach full I/O potential on only 4 channels. But isn't that just further evidence of the superior SAS technology?

    The SAS system may not be responsible for 100% of the performance difference but it is definitely the major contributing factor.
    There are no bugs - only undocumented features.
Sign In or Register to comment.