[rrd-developers] rrdtool 1.3 update slowness
Bernhard Fischer
rep.dot.nop at gmail.com
Tue Sep 4 15:56:26 CEST 2007
On Tue, Aug 28, 2007 at 04:28:13PM -0700, Evan Miller wrote:
>Disabling MMAP in 1.3 makes things worse:
>
>Create 10 rrds 1 c/s (0.00427 sdv) Update 10 rrds 22975 u/s (0.00013 sdv)
>Create 10 rrds 1 c/s (0.00428 sdv) Update 20 rrds 19045 u/s (0.00087 sdv)
>Create 20 rrds 1 c/s (0.00826 sdv) Update 40 rrds 12433 u/s (0.00205 sdv)
>Create 40 rrds 5 c/s (0.03187 sdv) Update 80 rrds 15 u/s (0.06627 sdv)
>Create 80 rrds 1 c/s (0.03359 sdv) Update 160 rrds 15 u/s (0.06745 sdv)
>Create 160 rrds 5 c/s (0.15173 sdv) Update 320 rrds 13 u/s (0.11210 sdv)
>
>I know very little about IO, so I have no idea what's going on. Let me know if
>there are other tests I can run, or if anyone can reproduce the behavior.
>Perftest is here:
>
>http://oss.oetiker.ch/rrdtool/pub/contrib/perftest.zip
perftest went into the repository a long time ago and is in examples/perftest.pl
I cannot really reproduce your report. I see:
perftest, 20k RRDs (100 per dir):
rrdtool-1.2.r1195:
100000 104 Up/s (0.05685 sdv)
100000 3126 Up/s (0.00428 sdv)
100000 12629 Up/s (0.00503 sdv)
100000 10722 Up/s (0.00997 sdv)
100000 11445 Up/s (0.00611 sdv)
100000 8878 Up/s (0.01648 sdv)
100000 11101 Up/s (0.00326 sdv)
100000 3965 Up/s (0.06022 sdv)
100000 3326 Up/s (0.07576 sdv)
100000 3374 Up/s (0.05249 sdv)
100000 3066 Up/s (0.08344 sdv)
rrdtool-1.3.r1196:
100000 19937 Up/s (0.00001 sdv)
100000 21399 Up/s (0.00002 sdv)
100000 22701 Up/s (0.00001 sdv)
100000 22742 Up/s (0.00001 sdv)
100000 60 Up/s (0.10694 sdv)
100000 21996 Up/s (0.00001 sdv)
100000 22448 Up/s (0.00002 sdv)
100000 21023 Up/s (0.00001 sdv)
100000 21835 Up/s (0.00001 sdv)
100000 22515 Up/s (0.00001 sdv)
100000 18455 Up/s (0.00001 sdv)
Where a very heavy I/O job was running for some time during the "60
Up/s" timeframe.
PS: Tobi recently changed perftest on trunk, i used the version from
trunk revision 1115 for the numbers above. rrdtool itself were the
revisions indicated above.
>
>Thanks,
>
>Evan
>
>On Tue, Aug 28, 2007 at 04:17:47PM -0700, Evan Miller wrote:
>> I ran Tobi's perftest.pl against RRDtool 1.2 and the current subversion
>> checkout. Updating a single RRD is noticeably faster under 1.3, but
>> performance grinds to a halt on my machine once I get up to 100 or so
>> RRDs. I consistently produce results similar to these:
>>
>>
>> RRDtool 1.2:
>>
>> Create 10 rrds 1 c/s (0.00117 sdv) Update 10 rrds 23288
>> u/s (0.00002 sdv)
>> Create 10 rrds 1 c/s (0.00062 sdv) Update 20 rrds 21125
>> u/s (0.00065 sdv)
>> Create 20 rrds 1 c/s (0.00125 sdv) Update 40 rrds 23385
>> u/s (0.00002 sdv)
>> Create 40 rrds 1 c/s (0.00259 sdv) Update 80 rrds 23150
>> u/s (0.00005 sdv)
>> Create 80 rrds 3 c/s (0.41032 sdv) Update 160 rrds 22688
>> u/s (0.00006 sdv)
>> Create 160 rrds 10 c/s (0.02412 sdv) Update 320 rrds 19940
>> u/s (0.00124 sdv)
>> Create 320 rrds 5 c/s (0.02459 sdv) Update 640 rrds 17601
>> u/s (0.00142 sdv)
>> Create 640 rrds 3 c/s (0.04289 sdv) Update 1280 rrds 10012
>> u/s (0.00432 sdv)
>> Create 1280 rrds 1 c/s (0.35445 sdv) Update 2560 rrds 4294
>> u/s (0.00614 sdv)
>> Create 768 rrds 1 c/s (0.05230 sdv) Update 3328 rrds 3626
>> u/s (0.00785 sdv)
>>
>>
>> RRDtool 1.3:
>>
>> Create 10 rrds 1 c/s (0.00771 sdv) Update 10 rrds 31434
>> u/s (0.00007 sdv)
>> Create 10 rrds 1 c/s (0.00421 sdv) Update 20 rrds 29388
>> u/s (0.00031 sdv)
>> Create 20 rrds 1 c/s (0.00937 sdv) Update 40 rrds 24646
>> u/s (0.00068 sdv)
>> Create 40 rrds 3 c/s (0.40489 sdv) Update 80 rrds 16121
>> u/s (0.00132 sdv)
>> Create 80 rrds 5 c/s (0.04052 sdv) Update 160 rrds 2000
>> u/s (0.00519 sdv)
>> Create 160 rrds 2 c/s (0.12109 sdv) Update 320 rrds 33
>> u/s (0.02963 sdv)
>>
>> $ uname -a
>> Linux yossarian 2.6.20-16-generic #2 SMP Thu Jun 7 20:19:32 UTC 2007
>> i686 GNU/Linux
>>
>> I didn't configure with any special options. Ideas?
>>
>> Evan
>>
>> _______________________________________________
>> rrd-developers mailing list
>> rrd-developers at lists.oetiker.ch
>> https://lists.oetiker.ch/cgi-bin/listinfo/rrd-developers
>
>_______________________________________________
>rrd-developers mailing list
>rrd-developers at lists.oetiker.ch
>https://lists.oetiker.ch/cgi-bin/listinfo/rrd-developers
>
More information about the rrd-developers
mailing list