Обсуждение: raid10 hard disk choice
Hello,
i have to buy a new server and in the budget i have (small) i have to select
one of this two options:
-4 sas 146gb 15k rpm raid10.
-8 sas 146gb 10k rpm raid10.
The server would not be only dedicated to postgresql but to be a file server,
the rest of options like plenty of ram and battery backed cache raid card are
done but this two different hard disk configuration have the same price and i am
not sure what it is better.
If the best option it is different for postgresql that for a file server i would
like to know too, thanks.
Regards,
Miguel Angel.
On Thu, 21 May 2009, Linos wrote: > i have to buy a new server and in the budget i have (small) i have to > select one of this two options: > > -4 sas 146gb 15k rpm raid10. > -8 sas 146gb 10k rpm raid10. It depends what you are doing. I think in most situations, the second option is better, but there may be a few situations where the reverse is true. Basically, the first option will only be faster if you are doing lots of seeking (small requests) in a single thread. As soon as you go multi-threaded or are looking at sequential scans, you're better off with more discs. Matthew -- Experience is what allows you to recognise a mistake the second time you make it.
On Thu, May 21, 2009 at 8:47 AM, Linos <info@linos.es> wrote: > Hello, > i have to buy a new server and in the budget i have (small) i have to > select one of this two options: > > -4 sas 146gb 15k rpm raid10. > -8 sas 146gb 10k rpm raid10. > > The server would not be only dedicated to postgresql but to be a file > server, the rest of options like plenty of ram and battery backed cache raid > card are done but this two different hard disk configuration have the same > price and i am not sure what it is better. > > If the best option it is different for postgresql that for a file server i > would like to know too, thanks. I would say go with the 10k drives. more space, flexibility (you can dedicate a volume to WAL), and more total performance on paper. I would also, if you can afford it and they fit, get two small sata drives, mount raid 1 and put the o/s on those. merlin
Matthew Wakeling wrote: > On Thu, 21 May 2009, Linos wrote: >> i have to buy a new server and in the budget i have (small) i >> have to select one of this two options: >> >> -4 sas 146gb 15k rpm raid10. >> -8 sas 146gb 10k rpm raid10. > > It depends what you are doing. I think in most situations, the second > option is better, but there may be a few situations where the reverse > is true. > > Basically, the first option will only be faster if you are doing lots > of seeking (small requests) in a single thread. As soon as you go > multi-threaded or are looking at sequential scans, you're better off > with more discs. > > Matthew > I agree. I think you would be better off with more disks I know from my own experience when I went from 8 73gb 15k drives to 16 73gb 15k drives I noticed a big difference in the amount of time it took to run my queries. I can't give you hard numbers but most of my queries take hours to run so you tend to notice when they finish 20-30 minutes sooner. The second option also doubles your capacity which in general is a good idea. It's always easier to slightly overbuild than try and fix a storage problem. Might I also suggest that you pick up at least one spare drive while you're at it. This may be somewhat of a tangent but it speaks to having a spare drive on hand. A word of warning for anyone out there considering the Seagate 1.5TB SATA drives (ST31500341AS). (I use them for an off-site backup of a backup array not pg) I'm going through a fiasco right now with these drives and I wish I had purchased more when I did. I built a backup array with 16 of these back in October and it works great. In October these drives shipped with firmware SD17. I needed to add another 16 drive array but the ST31500341AS drives that are currently shipping have a non-flashable CC1H firmware that will not work on high port count Adaptec cards (5XXXX) which is what I have. It is now impossible to find any of these drives with firmware compatible with my controller, trust me I spent a couple hours on the phone with Seagate. When I built the first array I bought a single spare drive. As soon as two drives die I'm going to be in the position of having to either scrap all of them or buy a new controller that will work with the new firmware. If I hadn't bought that extra drive the array would be dead as soon as one of the drives goes. My point is... if you have the means, buy at least one spare while you can. Bob
On Thu, 2009-05-21 at 10:25 -0400, Merlin Moncure wrote: > On Thu, May 21, 2009 at 8:47 AM, Linos <info@linos.es> wrote: > > Hello, > > i have to buy a new server and in the budget i have (small) i have to > > select one of this two options: > > > > -4 sas 146gb 15k rpm raid10. > > -8 sas 146gb 10k rpm raid10. > > > > The server would not be only dedicated to postgresql but to be a file > > server, the rest of options like plenty of ram and battery backed cache raid > > card are done but this two different hard disk configuration have the same > > price and i am not sure what it is better. > > > > If the best option it is different for postgresql that for a file server i > > would like to know too, thanks. > > I would say go with the 10k drives. more space, flexibility (you can > dedicate a volume to WAL), and more total performance on paper. I > would also, if you can afford it and they fit, get two small sata > drives, mount raid 1 and put the o/s on those. +1 on that. Joshua D. Drake > > merlin > -- PostgreSQL - XMPP: jdrake@jabber.postgresql.org Consulting, Development, Support, Training 503-667-4564 - http://www.commandprompt.com/ The PostgreSQL Company, serving since 1997
On Thu, May 21, 2009 at 8:34 AM, Robert Schnabel <schnabelr@missouri.edu> wrote: > the phone with Seagate. When I built the first array I bought a single > spare drive. As soon as two drives die I'm going to be in the position of > having to either scrap all of them or buy a new controller that will work > with the new firmware. If I hadn't bought that extra drive the array would > be dead as soon as one of the drives goes. > > My point is... if you have the means, buy at least one spare while you can. I'd go shopping for more spares on ebay now...
Matthew Wakeling wrote: > On Thu, 21 May 2009, Linos wrote: >> i have to buy a new server and in the budget i have (small) i have >> to select one of this two options: >> >> -4 sas 146gb 15k rpm raid10. >> -8 sas 146gb 10k rpm raid10. > > It depends what you are doing. I think in most situations, the second > option is better, but there may be a few situations where the reverse is > true. > > Basically, the first option will only be faster if you are doing lots of > seeking (small requests) in a single thread. As soon as you go > multi-threaded or are looking at sequential scans, you're better off > with more discs. Since you have to share the disks with a file server, which might be heavily used, the 8-disk array will probably be bettereven if you're doing lots of seeking in a single thread. Craig
On Thu, May 21, 2009 at 8:59 AM, Matthew Wakeling <matthew@flymine.org> wrote: > On Thu, 21 May 2009, Linos wrote: >> >> i have to buy a new server and in the budget i have (small) i have >> to select one of this two options: >> >> -4 sas 146gb 15k rpm raid10. >> -8 sas 146gb 10k rpm raid10. > > It depends what you are doing. I think in most situations, the second option > is better, but there may be a few situations where the reverse is true. One possible case of this - I believe that 15K drives will allow you to commit ~250 times per second (15K/60) vs. ~166 times per second (10K/60). If you have a lot of small write transactions, this might be an issue. ...Robert
On Thu, May 21, 2009 at 2:29 PM, Robert Haas <robertmhaas@gmail.com> wrote: > On Thu, May 21, 2009 at 8:59 AM, Matthew Wakeling <matthew@flymine.org> wrote: >> On Thu, 21 May 2009, Linos wrote: >>> >>> i have to buy a new server and in the budget i have (small) i have >>> to select one of this two options: >>> >>> -4 sas 146gb 15k rpm raid10. >>> -8 sas 146gb 10k rpm raid10. >> >> It depends what you are doing. I think in most situations, the second option >> is better, but there may be a few situations where the reverse is true. > > One possible case of this - I believe that 15K drives will allow you > to commit ~250 times per second (15K/60) vs. ~166 times per second > (10K/60). If you have a lot of small write transactions, this might > be an issue. But in a RAID-10 you aggreate pairs like RAID-0, so you could write 250(n/2) times per second on 15k where n=4 and 166(n/2) for 10k drives where n=8. So 500 versus 664... ? Or am I getting it wrong.
On 5/21/09 2:41 PM, "Scott Marlowe" <scott.marlowe@gmail.com> wrote: > On Thu, May 21, 2009 at 2:29 PM, Robert Haas <robertmhaas@gmail.com> wrote: >> On Thu, May 21, 2009 at 8:59 AM, Matthew Wakeling <matthew@flymine.org> >> wrote: >>> On Thu, 21 May 2009, Linos wrote: >>>> >>>> i have to buy a new server and in the budget i have (small) i have >>>> to select one of this two options: >>>> >>>> -4 sas 146gb 15k rpm raid10. >>>> -8 sas 146gb 10k rpm raid10. >>> >>> It depends what you are doing. I think in most situations, the second option >>> is better, but there may be a few situations where the reverse is true. >> >> One possible case of this - I believe that 15K drives will allow you >> to commit ~250 times per second (15K/60) vs. ~166 times per second >> (10K/60). If you have a lot of small write transactions, this might >> be an issue. > > But in a RAID-10 you aggreate pairs like RAID-0, so you could write > 250(n/2) times per second on 15k where n=4 and 166(n/2) for 10k drives > where n=8. So 500 versus 664... ? Or am I getting it wrong. From the original message: " The server would not be only dedicated to postgresql but to be a file server, the rest of options like plenty of ram and battery backed cache raid card are done but this two different hard disk configuration have the same price and i am not sure what it is better." So, with a write-back cache battery backed up raid card, xlog writes won't be an issue. > > -- > Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) > To make changes to your subscription: > http://www.postgresql.org/mailpref/pgsql-performance >
On Thu, May 21, 2009 at 5:41 PM, Scott Marlowe <scott.marlowe@gmail.com> wrote: > On Thu, May 21, 2009 at 2:29 PM, Robert Haas <robertmhaas@gmail.com> wrote: >> On Thu, May 21, 2009 at 8:59 AM, Matthew Wakeling <matthew@flymine.org> wrote: >>> On Thu, 21 May 2009, Linos wrote: >>>> >>>> i have to buy a new server and in the budget i have (small) i have >>>> to select one of this two options: >>>> >>>> -4 sas 146gb 15k rpm raid10. >>>> -8 sas 146gb 10k rpm raid10. >>> >>> It depends what you are doing. I think in most situations, the second option >>> is better, but there may be a few situations where the reverse is true. >> >> One possible case of this - I believe that 15K drives will allow you >> to commit ~250 times per second (15K/60) vs. ~166 times per second >> (10K/60). If you have a lot of small write transactions, this might >> be an issue. > > But in a RAID-10 you aggreate pairs like RAID-0, so you could write > 250(n/2) times per second on 15k where n=4 and 166(n/2) for 10k drives > where n=8. So 500 versus 664... ? Or am I getting it wrong. Well, that would be true if every write used a different disk, but I don't think that will be the case in practice. The WAL writes are very small, so often you'll have multiple writes even to the same block. But even if they're to different blocks they're likely to be in the same RAID stripe. ...Robert
On 5/21/09 3:05 PM, "Robert Haas" <robertmhaas@gmail.com> wrote: > On Thu, May 21, 2009 at 5:41 PM, Scott Marlowe <scott.marlowe@gmail.com> > wrote: >> On Thu, May 21, 2009 at 2:29 PM, Robert Haas <robertmhaas@gmail.com> wrote: >>> On Thu, May 21, 2009 at 8:59 AM, Matthew Wakeling <matthew@flymine.org> >>> wrote: >>>> On Thu, 21 May 2009, Linos wrote: >>>>> >>>>> i have to buy a new server and in the budget i have (small) i have >>>>> to select one of this two options: >>>>> >>>>> -4 sas 146gb 15k rpm raid10. >>>>> -8 sas 146gb 10k rpm raid10. >>>> >>>> It depends what you are doing. I think in most situations, the second >>>> option >>>> is better, but there may be a few situations where the reverse is true. >>> >>> One possible case of this - I believe that 15K drives will allow you >>> to commit ~250 times per second (15K/60) vs. ~166 times per second >>> (10K/60). If you have a lot of small write transactions, this might >>> be an issue. >> >> But in a RAID-10 you aggreate pairs like RAID-0, so you could write >> 250(n/2) times per second on 15k where n=4 and 166(n/2) for 10k drives >> where n=8. So 500 versus 664... ? Or am I getting it wrong. > > Well, that would be true if every write used a different disk, but I > don't think that will be the case in practice. The WAL writes are > very small, so often you'll have multiple writes even to the same > block. But even if they're to different blocks they're likely to be > in the same RAID stripe. Disk count and stripe size don't have much to do with it, the write cache merges write requests and the client (the wal log write) doesn't have to wait on anything. The RAID card can merge and order the writes, so it can go nearly at sequential transfer rate, limited more by other concurrent pressure on the raid card's cache than anything else. Since WAL log requests are sequential (but small) this provides huge gains and a large multiplier over the raw iops of the drive. > > ...Robert > > -- > Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) > To make changes to your subscription: > http://www.postgresql.org/mailpref/pgsql-performance >
On Thu, 21 May 2009, Scott Marlowe wrote: > But in a RAID-10 you aggreate pairs like RAID-0, so you could write > 250(n/2) times per second on 15k where n=4 and 166(n/2) for 10k drives > where n=8. So 500 versus 664... ? Or am I getting it wrong. Adding more spindles doesn't improve the fact that the disks can only commit once per revolution. WAL writes are way too fine grained for them to get split across stripes to improve the commit rate. -- * Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD
On Thu, 21 May 2009, Robert Schnabel wrote: > A word of warning for anyone out there considering the Seagate 1.5TB > SATA drives (ST31500341AS)...I'm going through a fiasco right now with > these drives and I wish I had purchased more when I did. Those drives are involved in the worst firmware debacle Seagate has had in years, so no surprise they're causing problems for you just like so many others. I don't think you came to the right conclusion for how to avoid this pain in the future though--buying more garbage drives isn't really satisfying. What you should realize is to never assemble a production server using newly designed drives. Always stay at least 6 months and at least one generation behind the state of the art. All the drive manufacturers right now are lucky if they can deliver a reliable 1TB drive, nobody has a reliable 1.5TB or larger drive yet. (Check out the miserable user ratings for all the larger capacity drives available right now on sites like newegg.com if you don't believe me) Right now, Seagate's 1.5TB drive is 7 months old, and I'd still consider it bleeding edge for server use. -- * Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD
Thanks for all the suggestions i will go with 8 10k disks, well 9 if you count the spare now that i am scared :) Regards, Miguel Angel.
Greg Smith wrote: > On Thu, 21 May 2009, Robert Schnabel wrote: >> A word of warning for anyone out there considering the Seagate 1.5TB >> SATA drives (ST31500341AS)...I'm going through a fiasco right now >> with these drives and I wish I had purchased more when I did. > I don't think you came to the right conclusion for how to avoid this > pain in the future though--buying more garbage drives isn't really > satisfying. No, the original drives I have work fine. The problem, as you point out, is that Seagate changed the firmware and made it so that you cannot flash it to a different version. > What you should realize is to never assemble a production server using > newly designed drives. I totally agree. I'm using these drives for an off-site backup of a backup. There is no original data on these. I needed the capacity. I was willing to accept the performance/reliability hit considering the $$/TB. Bob
On Fri, 22 May 2009, Robert Schnabel wrote: > No, the original drives I have work fine. The problem, as you point out, is > that Seagate changed the firmware and made it so that you cannot flash it to > a different version. The subtle point here is that whether a drive has been out long enough to have a stable firmware is very much a component of its overall quality and reliability--regardless of whether the drive works fine in any one system or not. The odds of you'll get a RAID compability breaking firmware change in the first few months a drive is on the market are painfully high. You don't have to defend that it was the right decision for you, I was just uncomfortable with the way you were extrapolating your experience to provide a larger rule of thumb. Allocated hot spares and cold spares on the shelf are both important, but for most people those should be a safety net on top of making the safest hardware choice, rather than as a way to allow taking excessive risks in what you buy. -- * Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD
On Fri, May 22, 2009 at 9:08 AM, Greg Smith <gsmith@gregsmith.com> wrote: > On Fri, 22 May 2009, Robert Schnabel wrote: > >> No, the original drives I have work fine. The problem, as you point out, >> is that Seagate changed the firmware and made it so that you cannot flash it >> to a different version. > > The subtle point here is that whether a drive has been out long enough to > have a stable firmware is very much a component of its overall quality and > reliability--regardless of whether the drive works fine in any one system or > not. The odds of you'll get a RAID compability breaking firmware change in > the first few months a drive is on the market are painfully high. Also keep in mind that 1.5 and 2TB drives that are out right now are all consumer grade drives, built to be put into a workstation singly or maybe in pairs. It's much less common to see such a change in server class drives, because the manufacturers know where they'll be used, and also because the server grade drives usually piggy back on the workstation class drives for a lot of their tech and bios, so the need for sudden changes are less common.
On Fri, 22 May 2009, Scott Marlowe wrote: > It's much less common to see such a change in server class drives This is a good point, and I just updated http://wiki.postgresql.org/wiki/SCSI_vs._IDE/SATA_Disks with a section about this topic (the last one under "ATA Disks"). -- * Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD
Greg Smith wrote: > On Fri, 22 May 2009, Scott Marlowe wrote: > >> It's much less common to see such a change in server class drives > > This is a good point, and I just updated > http://wiki.postgresql.org/wiki/SCSI_vs._IDE/SATA_Disks with a section > about this topic (the last one under "ATA Disks"). And I can confirm that point because the 1TB SAS drives (ST31000640SS) I just received to replace all the 1.5TB drives have the same firmware as the ones I purchased back in October. 60% more $$, 50% less capacity... but they work :-) Lesson learned.