Thread: server install recommendations?
hi everybody: in a couple of days i will reinstall an offline database server. It's a old HP Proliant DL580 G3 with three disks (147 GB each). Currently it has a debian Sarge in a RAID5 hardware array ( with HP Smart Array 6i, [RAID bus controller: Compaq Computer Corporation Smart Array 64xx (rev 01)] ). i googled a while and get to this page: http://www.powerpostgresql.com/PerfList , so my first question is, this document and the tips given hasn't been deprecated by now, right? then: -is raid5 the worst election in this scenario (three disks)? -which is the best (and with that i mean secure in first place and with more perfomance in second) possible configuration achievable with this three disks? -should i ask my boss to buy another disk? (it will be difficult, but if i can demonstrate It worth it...) i think i will "Separate the Transaction Log from the Database" with two RAID1 arrays (if they buy the new disk). is this a good way to go? it would be too bad if i put the logs in a disk without RAID? (only if i dont get the new disk...) our application (running on separate webserver) is of the type "OLTP", the server will be entirely dedicated to postgresql. i will configure a warm-standby server, so the WAL files will be forwarded to another server. thanks in advance! -- Roberto Scattini ___ _ ))_) __ )L __ ((__)(('(( ((_)
On Thu, Oct 16, 2008 at 7:22 AM, Roberto Scattini <roberto.scattini@gmail.com> wrote: > hi everybody: > > in a couple of days i will reinstall an offline database server. It's > a old HP Proliant DL580 G3 with three disks (147 GB each). Currently > it has a debian Sarge in a RAID5 hardware array ( with HP Smart Array > 6i, [RAID bus controller: Compaq Computer Corporation Smart Array 64xx > (rev 01)] ). Does that controller have battery backed cache? > i googled a while and get to this page: > http://www.powerpostgresql.com/PerfList , so my first question is, > this document and the tips given hasn't been deprecated by now, right? It's a little old but most of it holds true. > then: > -is raid5 the worst election in this scenario (three disks)? Yes > -which is the best (and with that i mean secure in first place and > with more perfomance in second) possible configuration achievable with > this three disks? either 2 disk mirror w hot spare or 3 disk mirror if the controller supports it. > -should i ask my boss to buy another disk? (it will be difficult, but > if i can demonstrate It worth it...) Yes you should. > i think i will "Separate the Transaction Log from the Database" with > two RAID1 arrays (if they buy the new disk). is this a good way to go? > it would be too bad if i put the logs in a disk without RAID? (only if > i dont get the new disk...) If the controller has battery backed cache, then go with a single RAID 1+0 unless it's really write heavy raid 1+0 will probably be a better choice. Do not use a single disk w no raid. > our application (running on separate webserver) is of the type "OLTP", > the server will be entirely dedicated to postgresql. i will configure > a warm-standby server, so the WAL files will be forwarded to another > server. You should benchmark with two RAID 1s and a single RAID1+0.
On Thu, 16 Oct 2008, Roberto Scattini wrote: > It's a old HP Proliant DL580 G3 with three disks (147 GB each). > Currently it has a debian Sarge in a RAID5 hardware array ( with HP > Smart Array 6i, [RAID bus controller: Compaq Computer Corporation Smart > Array 64xx (rev 01)] ). I think all of the 64xx controllers have a reasonable amount of battery-backed cache; with 3 disks you have a 6404 maybe? That's got 256MB of cache. The reason this is so important is that much of the advantage of having a separate disk for the WAL goes away with a good caching controller. Also, if you've only got a small number of database disks, you're not going to have the WAL as a bottleneck anyway. > -is raid5 the worst election in this scenario (three disks)? > -which is the best (and with that i mean secure in first place and > with more perfomance in second) possible configuration achievable with > this three disks? If I take "secure in the first place" to mean that you must be able to survive a disk failure, there are only two options here: -RAID5 with 3 disks -RAID1 pair with hotspare If your write load is low (you said OLTP which means it probably isn't) and you need as much space as possible the RAID5 might be a reasonable choice, but these are both relatively bad solutions. > -should i ask my boss to buy another disk? (it will be difficult, but > if i can demonstrate It worth it...) Having 4 disks in a RAID0+1 would be the ideal situation here for balancing performance and reliability. Just throw them all into one big volume and let the filesystem and controller balance everything out. It's unlikely you'll run into the WAL being the bottleneck if there's only two database disks and you have a caching controller. > i think i will "Separate the Transaction Log from the Database" with > two RAID1 arrays (if they buy the new disk). is this a good way to go? > it would be too bad if i put the logs in a disk without RAID? (only if > i dont get the new disk...) Do not consider the transaction logs to be an optional component less important than the database itself; if you lose them, you'll be hard pressed to recover from that. > our application (running on separate webserver) is of the type "OLTP", > the server will be entirely dedicated to postgresql. i will configure > a warm-standby server, so the WAL files will be forwarded to another > server. If you're keeping a warm-standby server around, the loss of a database disk might not be as big of a problem--you can keep that fairly up to date. Really depends on what the business guarantees required are. -- * Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD
Roberto Scattini wrote: > in a couple of days i will reinstall an offline database server. It's > a old HP Proliant DL580 G3 with three disks (147 GB each). Currently > it has a debian Sarge in a RAID5 hardware array ( with HP Smart Array > 6i, [RAID bus controller: Compaq Computer Corporation Smart Array 64xx > (rev 01)] ). Make sure you do a test installation first. I'm currently tinkering with a slightly older ProLiant and find that when I move from Debian "Sarge" to "Lenny" it won't recognise SMP. This is almost certainly a kernel issue and is probably something that won't get fixed, I suspect that 2.6 quite simply doesn't like the machine's SMBIOS structure. > -should i ask my boss to buy another disk? (it will be difficult, but > if i can demonstrate It worth it...) With this type of hardware you should have a good stock of identical disc drives before you start. You are very likely to find that the array firmware will reject any disc which isn't within 5 or 10% of the size of the existing ones, when hardware vendors rebadge (e.g. Seagate) disc drives they knock the apparent size down to a standard value. > our application (running on separate webserver) is of the type "OLTP", > the server will be entirely dedicated to postgresql. i will configure > a warm-standby server, so the WAL files will be forwarded to another > server. Again, make sure that you have spare compatible hardware- not ncessarily the same model but certainly using the same drive hardware, PSUs etc. If your boss won't spend the sort of money that these machines go for on eBay then your company isn't serious about running a service for its customers. -- Mark Morgan Lloyd markMLl .AT. telemetry.co .DOT. uk [Opinions above are the author's, not those of his employers or colleagues]