Results 1 to 10 of 10
  1. #1

    SATA or SAS for hosting game servers?

    I have past experience from running a GSP (2.5 years back).
    Most of the games i hosted were Q3-engine games, such as COD and SOF2 along with CS series.

    At that time, I never managed to get any quality hosting on servers with SATA drive(s) - I always ended up with SCSI drives and mostly Raid0. As I were hosting on 2U Dual Xenon machines, we did of course have a larger number of game servers on each machine - meaning more need of I/O on discs. But even when trying a smaller machine and less number of game servers - the disc was always a limiting factor.

    What's your experience on different drive types and also raid sets (R0)? We do of course have a lot of factors playing a role here, such as Punkbuster, game engine etc - but in general!
    Last edited by T1n3ll; 03-31-2010 at 04:12 AM.

  2. #2
    Join Date
    Jan 2010
    Location
    USA
    Posts
    2,145
    First of all, I would never use RAID0 in any kind of server. In RAID0, you can lose all of the data if one of the drives fails. I always use RAID10. With RAID10, you don't lose any data if one of the drives fail.

    Before you decide what drives to get, you may want to record some stats using MRTG to get an idea how many IOPS you need. Obviously SAS drives are faster than SATA drives, and SSD drives are faster than any hard drives. If you record some performance charts of the actual IOPS of reads and writes, then you can determine what type of drives to use. In many cases, you can get away with SATA drives in RAID10. If you need lots of IOPS, consider using SAS in RAID10 or SSD in RAID10.
    No Support Linux Hosting Bargain cPanel Hosting Experts Only
    We IGNORE the support questions, and pass the SAVINGS on to YOU!
    We also ignore questions about VPS Hosting

  3. #3
    Join Date
    Mar 2009
    Posts
    2,218
    Quote Originally Posted by T1n3ll View Post
    At that time, I never managed to get any quality hosting on servers with SATA drive(s) - I always ended up with SCSI drives and mostly Raid0.
    It's the disk speed that matters, not whether it is SAS or SATA.

    I think RAID 0 is generally a bad idea, as it can slow down disk I/O.

    Assume the stripe size is 64K bytes, and the application is reading 128K bytes. In non RAID, there should be one disk I/O reading 128K bytes; so the time will be

    S + R + 2T

    where S is the seek time, R is the rotational delay, and T is the transfer time for 64K bytes.

    With RAID 0, the time will be

    T + Max (S1+R1, S2+R2)

    Where (S1+R1) is seek time plus rotational delay for disk 1, (S2+R2) is seek time plus rotational delay for disk 2. Because some I/Os will be for smaller than the stripe size, the the seek armss for each disk can be anywhere.

    For I/Os greater than the stripe size, RAID 0 will take longer for seeking and positioning ... of the order of an extra 4 milliseconds on a 7.2K RPM disk. The parallel transfer only outweighs that when the transfer time exceeds 4 milliseconds - that is, for transfers of several hundred thousand bytes.

    *And* RAID 0 is using two disks instead of one, so for a busy system, I/O queues will be longer.

    RAID 0 was a good idea when disk transfer rates were 10 million bytes a second or less, so transfer time was a longer proportion of overall I/O time. But with modern disks transferring 100 million bytes a second or more, RAID 0 seems to me to make little sense compared with other methods of balancing I/O over multiple disks.

  4. #4
    There is no need for any raid when it comes to game server hosting. You will do fine with a single SATAII disk (7,200RPM).

  5. #5
    Join Date
    Dec 2005
    Location
    Berkshire, UK
    Posts
    3,070
    I've not hosted any gameservers for a while either, but I/O generally isn't that important once your daemon is sitting in memory. At least that was the case with CSS / TF2, Medal of Honour, Unreal Tournament servers etc.

    Definatley don't go for RAID0. It depends on the specification of the rest of the machine, but SATA disks will likely be more than adequete.
    SmartServerMangement (SSM)
    Specialists in SolusVM, VPS & cPanel Server Management

  6. #6
    Quote Originally Posted by PCS-Chris View Post
    I've not hosted any gameservers for a while either, but I/O generally isn't that important once your daemon is sitting in memory. At least that was the case with CSS / TF2, Medal of Honour, Unreal Tournament servers etc.

    Definatley don't go for RAID0. It depends on the specification of the rest of the machine, but SATA disks will likely be more than adequete.
    Correct, I/O isn't a concern. If you start getting larger nodes such as Dual 5520's you may want to consider 300GBRaptors or 4 x 7,200RPM's in Raid 10.

  7. #7
    Join Date
    Apr 2008
    Location
    United States, MI
    Posts
    766
    I/O on gameservers only matters when you:
    A) Change maps (though that's negligible)
    B) Write logs to disk (enough simultaneous writes can cause I/O wait)
    C) Use a database of sorts to store scores, admin levels, or things of that nature (negligible as well)
    D) Install mappacks, mods, or other "extras" via a control panel

    I/O is irrelevant just like PCS-Chris said when the gameserver is actually running in memory. Since all the libraries, textures, sounds, sprites, and so on are loaded into memory for that particular session.

    If you want better performance for your gameservers harddrive operations that I listed above, use a better filesystem (ext3/4). Raid is overkill on cost and power consumption.
    Steven Crothers
    No BS cloud engineer and Red Hat architect.

  8. #8
    Join Date
    Apr 2008
    Location
    United States, MI
    Posts
    766
    Quote Originally Posted by tim2718281 View Post
    It's the disk speed that matters, not whether it is SAS or SATA.

    I think RAID 0 is generally a bad idea, as it can slow down disk I/O.

    Assume the stripe size is 64K bytes, and the application is reading 128K bytes. In non RAID, there should be one disk I/O reading 128K bytes; so the time will be

    S + R + 2T

    where S is the seek time, R is the rotational delay, and T is the transfer time for 64K bytes.

    With RAID 0, the time will be

    T + Max (S1+R1, S2+R2)

    Where (S1+R1) is seek time plus rotational delay for disk 1, (S2+R2) is seek time plus rotational delay for disk 2. Because some I/Os will be for smaller than the stripe size, the the seek armss for each disk can be anywhere.

    For I/Os greater than the stripe size, RAID 0 will take longer for seeking and positioning ... of the order of an extra 4 milliseconds on a 7.2K RPM disk. The parallel transfer only outweighs that when the transfer time exceeds 4 milliseconds - that is, for transfers of several hundred thousand bytes.

    *And* RAID 0 is using two disks instead of one, so for a busy system, I/O queues will be longer.

    RAID 0 was a good idea when disk transfer rates were 10 million bytes a second or less, so transfer time was a longer proportion of overall I/O time. But with modern disks transferring 100 million bytes a second or more, RAID 0 seems to me to make little sense compared with other methods of balancing I/O over multiple disks.
    If your Raid0, your not using a 64kb stripe.... that's absurd. You would be using 128 just for the reasons you stated... Also your math is a bit off when it comes to the performance - your not accounting the A1, A2, B1, B2 methodology.
    Steven Crothers
    No BS cloud engineer and Red Hat architect.

  9. #9
    Join Date
    Sep 2008
    Location
    South Dallas, Texas
    Posts
    98
    Yeah i would go with RAID 10 man Dont make a mistake with RAID 0 as i have before, But you learn from your mistakes well at least i have

  10. #10
    Join Date
    Mar 2009
    Posts
    2,218
    Quote Originally Posted by Crothers View Post
    If your Raid0, your not using a 64kb stripe.... that's absurd. You would be using 128 just for the reasons you stated...
    In RAID 0, I/Os accessing two stripes can on average be slower than the equivalent I/O on non RAID.

    For typical 7.2K RPM drives, this will be the case for stripe sizes less than about a 500,000 bytes. Each disk will transfer data at about 100 million bytes per second, or 100,000 bytes per millisecond. So on a non-RAID system, transferring 500,000 bytes will take 5 milliseconds, compared with 2.5 milliseconds for a pair in RAID 0.

    The pair in RAID 0 will have to wait for two seeks plus rotational delays. These activities occur in parallel on the two drives; however, the completion time for the I/O depends on the slower of the two.

    For those who like analogies, consider the seek time and rotational delay being determined by the roll of dice. For a single disk, one die is rolled; the average delay is 3.5. For two disks, two dice are rolled, and the delay is whichever die gives the greater result. So the average delay is 4.5

    For a pair of 7.2K RPM disk in RAID 0, the average rotational delay will be 6ms instead of 4ms for a single non-RAID disk. And I estimate the average seek time for a full disk would in theory be about 14 ms instead of about 9ms; but in practice, there's a lot of locality in disk seeking, so the seek times on real systems are generally lower than manufacturers' figures would suggest.

    your math is a bit off when it comes to the performance - your not accounting the A1, A2, B1, B2 methodology.
    I said as the size of the I/O gets larger, RAID 0 reduces the performance deficit and eventually gets faster for I/Os larger than several hundred thousand bytes.

Similar Threads

  1. SAS v. SATA
    By dennisthompson in forum Computers and Peripherals
    Replies: 14
    Last Post: 03-26-2010, 08:51 AM
  2. SAS vs SATA II Performance
    By WireNine in forum Dedicated Server
    Replies: 6
    Last Post: 11-26-2008, 11:25 PM
  3. SAS / SATA compatibility
    By BostonGuru in forum Hosting Security and Technology
    Replies: 2
    Last Post: 11-20-2008, 01:29 AM
  4. SATA 7k RAID 1 vs. 2xSATA 10k vs. SATA+SAS
    By yosmc in forum Dedicated Server
    Replies: 24
    Last Post: 07-07-2008, 04:00 PM
  5. SATA vs. SAS
    By alphadog in forum Colocation and Data Centers
    Replies: 18
    Last Post: 09-21-2006, 08:02 AM

Posting Permissions

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