[Info-Ingres] Cache hit ratio 20% ummm is that good?
Paul White
shift7solutions at gmail.com
Wed Nov 14 23:01:48 CST 2007
I was horrified to find a cache hit ratio of under 20% on one of my customer
servers recently. Upon investigation the story is more interesting...
Fix Calls 169880982
Hits 28015058
Ratio 16%
I run a trace point DM420 every hour to collect useful stats on the cache.
Somewhere between 10am and 11am the FIX CALLS reset back to zero. Then later
the next day the HITS reset back to zero. It seems 2^32 is the maximum.
Here are the numbers...
FIX CALLS HIT GREADS
Mon Nov 12 07:42:01 2007 3949759223 3841806178 679286965
Mon Nov 12 08:42:00 2007 3950965028 3842871163 680044889
Mon Nov 12 09:42:00 2007 4055466875 3941987464 743076552
Mon Nov 12 10:42:00 2007 4210180730 4085219792 886492945
Mon Nov 12 11:42:00 2007 39901379 4205029337 942871883
Mon Nov 12 12:42:01 2007 43095584 4207918093 943989347
Mon Nov 12 13:42:00 2007 45047097 4209699466 944931919
...
Tue Nov 13 22:42:00 2007 138024523 4294163984 972833308
Tue Nov 13 23:42:00 2007 138314328 4294437209 973074381
Wed Nov 14 00:42:01 2007 139392284 393704 973820915
Wed Nov 14 01:42:00 2007 140144545 1098031 974425118
Wed Nov 14 02:42:00 2007 140941271 1853547 974792540
Wed Nov 14 03:42:01 2007 141392624 2276876 975180628
If I use only fix/hit to calculate the cache hit ratio I get weird numbers
after that point in time. Also note the cumulative figure gives 97% ratio
which is not useful for measuring performance based on recent activity.
FIX CALLS HIT ratio
3949759223 3841806178 97%
3950965028 3842871163 97%
4055466875 3941987464 97%
4210180730 4085219792 97%
39901379 4205029337 10539%
43095584 4207918093 9764%
...
45047097 4209699466 9345%
138024523 4294163984 3111%
138314328 4294437209 3105%
139392284 393704 0%
140144545 1098031 1%
140941271 1853547 1%
141392624 2276876 2%
Normally I use the delta between daily readings in order to calculate the
cache hit ratio. Now I think if the delta is negative I should just add
2^32.
DeltaFIX DeltaHIT ratio
3949759223 3841806178 97% (cumulative since the last restart)
1205805 1064985 88%
104501847 99116301 95%
154713855 143232328 93%
124687945 119809545 96%
3194205 2888756 90%
...
1951513 1781373 91%
92977426 84464518 91%
289805 273225 94%
1077956 923791 86%
752261 704327 94%
796726 755516 95%
451353 423329 94%
Have I bored you all to tears? The server is rebooted monthly so the
problem only shows up after some very heavy processing (eg a process went
crazy)
ps.. Would anyone like to comment on my GREAD numbers? I think they are too
high for a transaction based database.
cheers
Paul
More information about the Info-Ingres
mailing list