[jira] [Updated] (CASSANDRA-7980) cassandra-stress should support partial clustering column generation
[ https://issues.apache.org/jira/browse/CASSANDRA-7980?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sylvain Lebresne updated CASSANDRA-7980: Assignee: (was: Branimir Lambov) > cassandra-stress should support partial clustering column generation > > > Key: CASSANDRA-7980 > URL: https://issues.apache.org/jira/browse/CASSANDRA-7980 > Project: Cassandra > Issue Type: Improvement > Components: Testing >Reporter: Benedict >Priority: Minor > > cassandra-stress generates its data randomly, in tiers, so that we can scroll > through the partitions it generates without having to generate their > entirety. The problem is that to support very large partitions (important for > benchmarking certain cases, and acceptance testing) we have to have a large > number of clustering columns - generally more than we would otherwise have, > which changes the performance characteristics. We should effectively split > each clustering column into a number of byte-ranges that become tiers for > visitation. The only real complexity here is in obeying the size/count > distribution range specified, which would be difficult for exponential > distributions, however we could require the user specify the ranges, and > distributions for each range, upfront. We could even treat them exactly like > other column specifications, but as sub-specs within a given column in the > yaml. Or, we could simply accept that we imperfectly follow the distribution > in these situations. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Updated] (CASSANDRA-7980) cassandra-stress should support partial clustering column generation
[ https://issues.apache.org/jira/browse/CASSANDRA-7980?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Branimir Lambov updated CASSANDRA-7980: --- Component/s: Testing > cassandra-stress should support partial clustering column generation > > > Key: CASSANDRA-7980 > URL: https://issues.apache.org/jira/browse/CASSANDRA-7980 > Project: Cassandra > Issue Type: Improvement > Components: Testing >Reporter: Benedict >Assignee: Branimir Lambov >Priority: Minor > > cassandra-stress generates its data randomly, in tiers, so that we can scroll > through the partitions it generates without having to generate their > entirety. The problem is that to support very large partitions (important for > benchmarking certain cases, and acceptance testing) we have to have a large > number of clustering columns - generally more than we would otherwise have, > which changes the performance characteristics. We should effectively split > each clustering column into a number of byte-ranges that become tiers for > visitation. The only real complexity here is in obeying the size/count > distribution range specified, which would be difficult for exponential > distributions, however we could require the user specify the ranges, and > distributions for each range, upfront. We could even treat them exactly like > other column specifications, but as sub-specs within a given column in the > yaml. Or, we could simply accept that we imperfectly follow the distribution > in these situations. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Updated] (CASSANDRA-7980) cassandra-stress should support partial clustering column generation
[ https://issues.apache.org/jira/browse/CASSANDRA-7980?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Benedict updated CASSANDRA-7980: Issue Type: Improvement (was: Bug) cassandra-stress should support partial clustering column generation Key: CASSANDRA-7980 URL: https://issues.apache.org/jira/browse/CASSANDRA-7980 Project: Cassandra Issue Type: Improvement Reporter: Benedict Assignee: Branimir Lambov Priority: Minor cassandra-stress generates its data randomly, in tiers, so that we can scroll through the partitions it generates without having to generate their entirety. The problem is that to support very large partitions (important for benchmarking certain cases, and acceptance testing) we have to have a large number of clustering columns - generally more than we would otherwise have, which changes the performance characteristics. We should effectively split each clustering column into a number of byte-ranges that become tiers for visitation. The only real complexity here is in obeying the size/count distribution range specified, which would be difficult for exponential distributions, however we could require the user specify the ranges, and distributions for each range, upfront. We could even treat them exactly like other column specifications, but as sub-specs within a given column in the yaml. Or, we could simply accept that we imperfectly follow the distribution in these situations. -- This message was sent by Atlassian JIRA (v6.3.4#6332)