Re: Vacuum problem - Mailing list pgsql-general

On 5/13/2013 7:10 PM, S H wrote:
> My disk is utilized by many other components, thus do we have minimum
> recommendation my postgres to have sufficient speed. Current
> perfomance of my disk is around 1-5MB/sec. Is it sufficient?

how are you measuring this?   thats painfully slow by today's standards,
even my desktop SATA drives can sustain well over 100MB/second on
sequential read or write

what counts in a database server like postgres is NOT the sequential
transfer speed, instead its the random IO operations/second.   I'm
benchmarking a 8 disk RAID10 right now and seeing around 2000-4000
write/sec and as high as 1000-2400 read/sec during this TPC-B style
transaction benchmark.    iostat -xm during this operation looks like...

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
            3.34    0.00    1.93   39.43    0.00   55.31

Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s wMB/s avgrq-sz
avgqu-sz   await  svctm  %util
sdb               0.00     0.10  153.50 4901.30     1.89 40.69
17.25   211.50   42.57   0.20 100.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           10.95    0.00    3.98   36.35    0.00   48.72

Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s wMB/s avgrq-sz
avgqu-sz   await  svctm  %util
sdb               0.00     0.10  583.40 2955.40     7.10 28.75
20.75   215.77   63.01   0.28  99.97

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           44.36    0.00   14.28   24.61    0.00   16.75

Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s wMB/s avgrq-sz
avgqu-sz   await  svctm  %util
sdb               0.00     1.10 2377.30 1062.90    29.23 29.97
35.24    29.53    8.59   0.29  99.95

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           48.39    0.00   16.45   21.18    0.00   13.98

Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s wMB/s avgrq-sz
avgqu-sz   await  svctm  %util
sdb               0.00     0.00 2564.30 1182.40    31.15 32.42
34.75    28.81    7.69   0.27  99.97

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           44.59    0.00   15.02   25.38    0.00   15.02

Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s wMB/s avgrq-sz
avgqu-sz   await  svctm  %util
sdb               0.00     0.10 2346.10 1097.50    28.66 29.28
34.46    29.58    8.60   0.29  99.98

(these are 10 second averages)

in general, the solution to more storage performance for a database
server is to use more disks in a raid10 configuration,  I have some
raids that are a many as 20 disks, dedicated to database use (everything
else on the server uses other storage).  These are all 15000rpm SAS2
server drives, on a raid controller with 1GB flash-backed write-back cache.




--
john r pierce                                      37N 122W
somewhere on the middle of the left coast