On Thursday 24 June 2010 14:53:57 Matthew Wakeling wrote: > On Thu, 24 Jun 2010, Janning wrote: > > We have a 12 GB RAM machine with intel i7-975 and using > > 3 disks "Seagate Barracuda 7200.11, ST31500341AS (1.5 GB)" > > Those discs are 1.5TB, not 1.5GB. sorry, my fault. > > One disk for the system and WAL etc. and one SW RAID-0 with two disks for > > postgresql data. Our database is about 24GB. > > Beware of RAID-0 - make sure you can recover the data when (not if) a disc > fails. oh sorry again, its a raid-1 of course. shame on me. > > Our munin graph reports at 9:00 a clock writes of 3000 blocks per second > > and reads of about 1000 blocks per second on our disk which holds the > > data directories of postgresql (WAL are on a different disk) > > > > 3000 blocks ~ about 3 MB/s write > > 1000 blocks ~ about 1 MB/s read > > > > At the same time we have nearly 50% CPU I/O wait and only 12% user CPU > > load (so 4 of 8 cpu cores are in use for io wait) > > Not quite sure what situation you are measuring these figures under. > However, as a typical figure, let's say you are doing random access with > 8kB blocks (as in Postgres), and the access time on your drive is 8.5ms > (as with these drives). > > For each drive, you will be able to read/write approximately 8kB / > 0.0085s, giving 941kB per second. If you have multiple processes all doing > random access, then you may be able to utilise both discs and get double > that. So with your calculation I have a maximum of 2MB/s random access. So i really need to upgrade my disk configuration! > > Of course, writing large chunks is quite a different usage pattern. But I > > am wondering that writing 3MB/s and reading 1 MB/s seams to be a limit if > > i can run a test with 89 MB/s writing and 110MB/s reading. > > That's quite right, and typical performance figures for a drive like that. thanks for your help. kind regards Janning > Matthew > > -- > Don't criticise a man until you have walked a mile in his shoes; and if > you do at least he will be a mile behind you and bare footed. -- Sent via pgsql-performance mailing list (pgsql-performance@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
