[ https://issues.apache.org/jira/browse/CASSANDRA-1520?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12912288#action_12912288 ]
Brandon Williams commented on CASSANDRA-1520: --------------------------------------------- Committed as is, since multiget was fairly poor. Number of threads as a stopgap probably isn't too bad, but ultimately I think providing a config option for the number of keys is best, so the user knows exactly what's going on. > stress.py's multiget option sends increasingly inefficient queries as more > test data is inserted > ------------------------------------------------------------------------------------------------ > > Key: CASSANDRA-1520 > URL: https://issues.apache.org/jira/browse/CASSANDRA-1520 > Project: Cassandra > Issue Type: Bug > Affects Versions: 0.7 beta 1 > Reporter: Nate McCall > Priority: Minor > Fix For: 0.7 beta 2 > > Attachments: 1520.patch > > > MultiGetter's key list sizes should be broken up better for more efficient > queries. Setting an initial value that breaks up the key list into N sub > lists (where N is the number of threads) yielded more efficient queries. (The > choice of thread count here was a stop-gap for demonstration purposes. End > result should probably be chunk-size config option with a sane default). > Pre patch: > --- > python stress.py -o multiget -t 25 -n 250000 -c 5 > total,interval_op_rate,interval_key_rate,avg_latency,elapsed_time > 6,0,6000,8.6109764576,10 > 10,0,4000,18.6666852832,20 > 17,0,7000,27.4705835751,30 > 23,0,6000,36.6091703971,41 > 25,0,2000,41.8415510654,42 > Post patch: > --- > python mstress.py -o multiget -t 25 -n 250000 -c 5 > total,interval_op_rate,interval_key_rate,avg_latency,elapsed_time > 172,17,6880,1.44215503127,10 > 314,14,5680,1.8667214538,20 > 466,15,6080,1.69888155084,31 > 624,15,6320,1.55442555947,41 > 625,0,40,0.0914790630341,41 -- This message is automatically generated by JIRA. - You can reply to this email to add a comment to the issue online.