Page 1 of 2 12 LastLast
Results 1 to 25 of 32
  1. #1
    Join Date
    Nov 2001
    Location
    Toronto
    Posts
    220

    Raptor 10k SATA drives vs. SCSI 10k

    I'm thinking of buying a bunch of Raptor 10k drives for a database intesive server because of cost factors. I looked at a few reviews on the Raptor SATA drives which deliver 10k RPM, not bad. Is anyone out there using these drives? I'll save a whole bunch if I opt for these drives as opposed to a SCSI setup.

    Also how much performance gain is there from placing the database on different drives than the OS?

  2. #2
    I'm usualy very PRO sata based on price/performance. But only if its 1 or 2 disks in the system (and only if you use quality ones, like seagate, avoid anything with a 1 year waranty as that means its not even rated for datacenter use!).

    You said buying a 'bunch' so im going to presume you plan a raid 5 type configuration probably with 4 or 5 of these drives. In that case I would definately go SCSI, instead of SATA. Scsi's performance is not really that big a difference until you hit multiple disks with simultaneous read/write spread across all drives (which is what raid 5, and 4 or 5 drives is going to do). Scsi will be probably as much as 50% faster in fact in that situation. (Main reason being that SCSI has a cpu on the card that off loads a lot of the work and can handle sequential across many drives at once at high speed, where as sata will put strain on your main cpu/IO buffers and is not well tuned to write to many drives at once). There are some newer SATA II cards that do a better job, but I haven't played with them.

    If you just were doing 2 drives, I'd go SATA as the performance there should be about 10-15% at most on non stop intenstive usage (And if the box isn't pushing max usage 24 x 7 then sata should be almost identical).

    As for 'what drive' stores the OS vs the database, if you are doing raid 5 it wont matter, but it really depends how you plan to string together 'this bunch of drives', i.e. Linux LVM software partitions? I'd go raid 5, and put the 'boot partition' on a non raid 5 partition to help on boot ups in case there is ever a problem.
    Colohouse | P: 877-539-4638 | E: hello@colohouse.com
    Dedicated Servers | Cloud VPS | Colocation | Web Hosting | Public & Private Cloud | IaaS & DRaaS
    Data Centers: New York | Chicago | Miami | Colorado | California | Amsterdam

  3. #3
    Join Date
    Nov 2001
    Location
    Toronto
    Posts
    220
    Thank you for that reply! Yes, I defenitely need a RAID setup and I was thinking a RAID 0+1 for the OS and a RAID 5 for the database drives. But you're saying with a RAID 5 it won't matter if the OS is on separate drives. How's that? Wouldn't that result in less seeks on the database drives? Especially since there will be lots of logging from Apache and the web site itself.

  4. #4
    In Raid 5, all drives 'merge' into 1. Every read/write is spread out across all drives already to balance it (and help on performance) . So in essence it wont matter, you cant specify on raid 5 "write my logs to drive 1, and database to drive 2". Raid 5 is just 'all of them'. It is possible to configure raid 5 to just use 4 of your 5 hard disks, and then install the OS on a seperate disk - would i suggest it? Not if i was putting 5 x 10k rpm drives in there, as you are wasting performance in my opinion. But i would put the boot partition on a non raid 5 (if i was using SOFTWARE raid), to make life easier in case trouble ever hits and corrupts things. But not the main OS and log folders, those could go in the raid 5 just fine.
    Colohouse | P: 877-539-4638 | E: hello@colohouse.com
    Dedicated Servers | Cloud VPS | Colocation | Web Hosting | Public & Private Cloud | IaaS & DRaaS
    Data Centers: New York | Chicago | Miami | Colorado | California | Amsterdam

  5. #5
    Last I checked, Raptor drives are just as expensive as SCSI 10k drives, no?

    A DB on RAID5 will have write performance issues(noticable), fyi. The optimal DB setup would be RAID1 the OS, RAID1+0 the DB MDF and RAID1 the Log file. Install the DB executable on the OS drive, or it's own RAID1 if you have the space. Although if you have the space to put the DB executable on it's own RAID1, I'd use that space for the OS swap file, and still stick the DB EXE on the OS RAID.

    Chip-

  6. #6
    Join Date
    Apr 2004
    Posts
    1,834
    Quote Originally Posted by carguy84
    Last I checked, Raptor drives are just as expensive as SCSI 10k drives, no?
    yes and 1/3 the performance and reliablity. SATA is a consumer item, SCSI is a comercial product.
    Ray Womack @ atOmicVPS LTD
    Linux & Windows Cloud Hosting Solutions Powered by OnApp
    Fully Managed [Shared][Reseller][Cloud VPS] [Dedicated]
    Featuring the atOmicSTACK ● Speed ● Performance ● Reliability

  7. #7
    Join Date
    Jun 2003
    Location
    San Francisco
    Posts
    623
    Quote Originally Posted by PSFServers
    yes and 1/3 the performance and reliablity. SATA is a consumer item, SCSI is a comercial product.
    The Raptor SATA is essential the same as their SCSI drives except the interface. The reliability is rated the same.

    Performance is for everyone to argue about.

  8. #8
    Join Date
    Jun 2002
    Location
    PA, USA
    Posts
    5,143
    I would like to see where Raptor is rated as the same in reliability and performance then SCSI drives.
    Fluid Hosting, LLC - Enterprise Cloud Infrastructure: Cloud Shared and Reseller, Cloud VPS, and Cloud Hybrid Server

  9. #9
    we've used 16x74GB 10k RPM raptors with a 16port 3ware 9600 series cards and they sustain around 450-500mbps of throughput for file downloads etc.

    no complaints here..

    this is running on a 800FSB 3.06/2mb cache single cpu xeon NFS server. i find it that you need a quality controller.

    while scsi is obviously "ideal" the cost per MB and for my application dont warrant the high cost.
    Tim M./ Krypt.com a division of VPLS,Inc.
    Find us on Facebook!!
    Website: www.krypt.com / Twitter: @KryptServers
    Office: 866-616-9099 x9090 Fax: 866-599-9593

  10. #10
    Join Date
    Dec 2004
    Location
    San Francisco, CA
    Posts
    1,912
    Yeah, I doubt Raptor SATA disks can match the reliability of a SCSI disk..

    Also, don't forget, SATA uses your CPU as a controller. SCSI has its own controller. Heavy disk activity will put more load on the processors on a SATA system than on a SCSI system...
    init.me - Build, Share & Embed

    JodoHost.com - Windows VPS Hosting, ASP.NET and SQL Server Hosting
    8th year in Business, 200+ Servers. Microsoft Gold Certified Partner

  11. #11
    Join Date
    Apr 2005
    Location
    Jacksonville, FL
    Posts
    981
    Quote Originally Posted by krypttim
    we've used 16x74GB 10k RPM raptors with a 16port 3ware 9600 series cards and they sustain around 450-500mbps of throughput for file downloads etc.
    That's not surprising at all. Because you mentioned 3ware, I'm assuming some sort of RAID. These cards are capable of upwards of 450-500 megabytes per second on reads, 8 times faster than the figure of 450-500 megabits per second. The manufacturer conservatively rates them at over 400 Mbytes/sec (see http://www.3ware.com/products/serial_ata9000.asp).

    Performance-wise, 10k RPM raptors perform better than comparable SCSI drives on sustained reads, but they apparently suck on multiple simultaneous IO's. This review really tells the whole story.

    As for reliability, I've never had any failures, but I know a few people who have. At first glance they certainly don't seem as reliable as a Seagate SCSI, but probably on par with the MTBF of a typical Quantum (Maxtor) Atlas or IBM (Hitachi) drive.

    When building a RAID array, my money is on SATA Raptors, hands-down. Remember what RAID is - Redundant Array of INEXPENSIVE Disks. While the drives are not that much cheaper than SCSI, there's alot of value in integrated SATA RAID on most new motherboards, which typically don't cost much more than their non-RAID counterparts.

    Now selling BigVPS's!
    Jacksonville Colocation and dedicated servers by colo4jax
    We are *not* a reseller. We own our servers, switches, routers and racks.

  12. #12
    Join Date
    Jun 2003
    Location
    San Francisco
    Posts
    623
    The Raptor is rated MTBF at 1.2 million hours, same as SCSI drives. Look up the spec and you will see it. It uses the same mechanism as enterprise SCSI drives, except the interface. That's why it is so expensive comparing to regular SATA drives.

    SATA drives will out-perform SCSI for sequencial read/write in large RAID, due to each drive has its own channel. SCSI would max the channel limit after 5 to 6 drives.

    I would still use SCSI for DB. It is a lot of small random read/write and the bandwidth limit is not an issue.

  13. #13
    Join Date
    Dec 2004
    Location
    San Francisco, CA
    Posts
    1,912
    Your Raptor Drive is only priced $40 less than a Maxtor 10K SCSI Drive.. Given the falling prices of SCSI, I see no reason to go for SATA..

    Also as clearly shown by that review, SCSIs will outperform any SATA drive in a server environment.. according to that review, by a whopping 65%.. That's huge IMHO. In a desktop environment, the Rapter drive outperforms a SCSI drive by a very small margin....

    A desktop environment doesn't have too many simultaneous file reads/writes... as we all know. A server environment is going to have tons of simultaneous read/writes and that is where the SCSI's inbuilt controller gives it such a huge advantage

    Spending a little more on SCSI disks means better performance and reduced server loads, as well as better disk reliability. We use SCSI RAID5 disk arrays for our servers and I honestly see it more economical to run than SATA or Raptor drives
    init.me - Build, Share & Embed

    JodoHost.com - Windows VPS Hosting, ASP.NET and SQL Server Hosting
    8th year in Business, 200+ Servers. Microsoft Gold Certified Partner

  14. #14
    Join Date
    Jun 2003
    Location
    San Francisco
    Posts
    623
    Not only the Raptor is not much cheaper than 10k SCSI's, the RAID card will end up more expensive than SCSI RAID. A good 2 channel SCSI RAID costs around $400, and support 24 drives. A good (3ware) 12 SATA RAID card cost $600 and only support 12 drives. I am comparing PCI-X to PCI-X.

  15. #15
    What do people make of this article?
    http://www.pugetsystems.com/articles.php?id=19

  16. #16
    Join Date
    Jan 2003
    Location
    Chicago, IL
    Posts
    6,957
    Quote Originally Posted by tical
    Performance-wise, 10k RPM raptors perform better than comparable SCSI drives on sustained reads, but they apparently suck on multiple simultaneous IO's. This review really tells the whole story.
    That doesn't tell the whole story as it doesn't cover any of the improvements in the new 150GB Raptor, it now has increase I/O as it has NCQ instead of TCQ. They have also increased both the throughput speed and the seek times. The new 150GB drives are a good bit faster than the old 74GB drives. Those 74GB ones are what, ttwo years old now?
    Karl Zimmerman - Founder & CEO of Steadfast
    VMware Virtual Data Center Platform

    karl @ steadfast.net - Sales/Support: 312-602-2689
    Cloud Hosting, Managed Dedicated Servers, Chicago Colocation, and New Jersey Colocation

  17. #17
    Join Date
    Jun 2002
    Posts
    41
    http://www.storagereview.com/article...500ADFD_6.html

    this review compares the newer 150gb raptors with ncq with older generation 10k scsi's in a multi user environment. from experience, in a web server environment, i can vouch that 10k scsi's are much faster than a 10k 150 raptor with ncq enabled. even when pitched against a raid1 10k 150 raptor, i would go with the single 10k scsi anyday.
    Last edited by jusunlee; 04-14-2006 at 04:54 PM.

  18. #18
    Join Date
    Oct 2002
    Posts
    227
    Quote Originally Posted by jusunlee
    http://www.storagereview.com/article...500ADFD_6.html

    this review compares the newer 150gb raptors with ncq with older generation 10k scsi's in a multi user environment. from experience, in a web server environment, i can vouch that 10k scsi's are much faster than a 10k 150 raptor with ncq enabled. even when pitched against a raid1 10k 150 raptor, i would go with the single 10k scsi anyday.
    This is very true. I've tried raptors and, in the real world, 10K SCSI wins hands down.

  19. #19
    In terms of speed (forgetting about risk/data loss/life span etc), which one would you choose:

    1) 10k Raptor NCQ x 4 in RAID 0 setup

    OR

    2) 10 k SCSI x 1, non-RAID

    ???

  20. #20
    Quote Originally Posted by riverpast
    The Raptor is rated MTBF at 1.2 million hours, same as SCSI drives.
    And how often do we see drives lasting almost 137 years?
    Silicon Valley Web Hosting - Bay Area Bare-Metal and 1U to Full Rack Colocation

    www.svwh.net

  21. #21
    Join Date
    Aug 2005
    Location
    Seattle, WA
    Posts
    1,169
    Quote Originally Posted by Erwin
    In terms of speed (forgetting about risk/data loss/life span etc), which one would you choose:

    1) 10k Raptor NCQ x 4 in RAID 0 setup

    OR

    2) 10 k SCSI x 1, non-RAID

    ???
    In webhosting, It'd be much more beneficial to run a configuration like RAID 5.

  22. #22
    I think he actually (correct me if I am wrong) meant Raid 10 for #1.

    And for a database and 4 drives, I would go with RAID 10 over RAID 5. Much faster and still gives you good fail-over unless you are on a serious vacation and couldn't catch a drive or even maybe two failing.

  23. #23
    Quote Originally Posted by Erwin
    In terms of speed (forgetting about risk/data loss/life span etc), which one would you choose:

    1) 10k Raptor NCQ x 4 in RAID 0 setup

    OR

    2) 10 k SCSI x 1, non-RAID

    ???
    I mean RAID 0 without any redundancy - if you had to just look at SPEED (without anything else) which would be faster?

  24. #24
    That is a good question, but assuming you are using 4 drives in RAID 0, if you lost one, wouldn't you loose the entire array? Raid 10 lets you have two RAID 0 setups that are mirrored to each other in a RAID 1 setup. If you lose 1 drive you don't kill all 4 then. I think that is how it works out.

  25. #25
    I know of the risks, but I'm just wondering if using 4 Raptors in a RAID 0 array would actually outrun a 10k single SCSI.

Page 1 of 2 12 LastLast

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •