tag:blogger.com,1999:blog-24359421.post3535215797015936276..comments2023-04-29T09:47:48.006+02:00Comments on PrimeBase XT: Improving PBXT DBT2 PerformanceUnknownnoreply@blogger.comBlogger11125tag:blogger.com,1999:blog-24359421.post-7695635045430956652009-03-16T09:56:00.000+01:002009-03-16T09:56:00.000+01:00Hi Peter,Must be a bug. I have been using --databa...Hi Peter,<BR/><BR/>Must be a bug. I have been using --database=[db-name] to indicate the database.<BR/><BR/>Note that by default the dbt2 build has the STANDALONE switch enabled. This makes things a lot simpler.<BR/><BR/>The driver calls the database directly and does not use the "terminal server".<BR/><BR/>This makes it possible to start the test without using the script. for example:<BR/><BR/>mkdir output<BR/>mkdir output/report<BR/>mkdir output/driver<BR/>src/driver -l 30 -wmin 1 -wmax 5 -w 5 -sleep 0 -tpw 8 -outdir output/driver -ktd 0 -ktn 0 -kto 0 -ktp 0 -kts 0 -ttd 0 -ttn 0 -tto 0 -ttp 0 -tts 0 --socket=/tmp/mysql.sock --database=dbt2_pbxt_w5 -dbc 16; src/analyze_dbt_2 output ; cat output/report/mix.log.summaryPaul McCullaghhttps://www.blogger.com/profile/10476275147692948727noreply@blogger.comtag:blogger.com,1999:blog-24359421.post-32890590881013888342009-03-16T09:40:00.000+01:002009-03-16T09:40:00.000+01:00hi Paul, I tried to build and run dbt2 on launchpa...hi Paul, <BR/>I tried to build and run dbt2 on launchpad(https://code.launchpad.net/~dbt2-core-dev/dbt2/trunk), and found that script/run_workload.sh call src/driver with wrong argument.<BR/>The argument: " -d ${CLIENT_HOSTNAME} " is rejected by src/driver.<BR/>How to fix the problem, Paul? <BR/>Thank you.Unknownhttps://www.blogger.com/profile/03840850753871267265noreply@blogger.comtag:blogger.com,1999:blog-24359421.post-87370677324862294422009-03-07T00:12:00.000+01:002009-03-07T00:12:00.000+01:00Hi Ronald!You are absolutely right about DBT2. Its...Hi Ronald!<BR/><BR/>You are absolutely right about DBT2. Its an old benchmark which doesn't have all that much to do with most applications being run by MySQL today.<BR/><BR/>With your field experience you know this better than me.<BR/><BR/>But I have been told it is an important industry standard so, "I just got to do it" :)<BR/><BR/>Regarding SSDs, you are right, there is a huge difference between drives today.<BR/><BR/>The one I am using is a fairly old (> 1 year) 32GB Mtron drive. I don't know the exact characteristics but the newer drives, for example from Intel, are a lot faster.Paul McCullaghhttps://www.blogger.com/profile/10476275147692948727noreply@blogger.comtag:blogger.com,1999:blog-24359421.post-61792629565825497732009-03-06T23:53:00.000+01:002009-03-06T23:53:00.000+01:00Hi Ryan,innodb_flush_log_at_trx_commit was set to ...Hi Ryan,<BR/><BR/>innodb_flush_log_at_trx_commit was set to 1.<BR/><BR/>Both PBXT and InnoDB are flushing every commit in the test, i.e. full durability for all transactions.<BR/><BR/>The TPMs are quite high, largely due to the SSD's high flush rate.Paul McCullaghhttps://www.blogger.com/profile/10476275147692948727noreply@blogger.comtag:blogger.com,1999:blog-24359421.post-59289458276520646762009-03-06T23:50:00.000+01:002009-03-06T23:50:00.000+01:00Hi Roland & Giuseppe!Thanks to both of you for...Hi Roland & Giuseppe!<BR/><BR/>Thanks to both of you for your support.<BR/><BR/>I look forward to meeting again at the conference...Paul McCullaghhttps://www.blogger.com/profile/10476275147692948727noreply@blogger.comtag:blogger.com,1999:blog-24359421.post-84328128518516256492009-03-06T23:43:00.000+01:002009-03-06T23:43:00.000+01:00Hi Erkan,About the HA solution. The first step is ...Hi Erkan,<BR/><BR/>About the HA solution. The first step is asynchronous engine level replication.<BR/><BR/>We hope to be able to show something in this area by the 3rd quarter this year.<BR/><BR/>So we still have a way to go ... :)Paul McCullaghhttps://www.blogger.com/profile/10476275147692948727noreply@blogger.comtag:blogger.com,1999:blog-24359421.post-14672289676088228042009-03-06T21:19:00.000+01:002009-03-06T21:19:00.000+01:00A 100%+ improvement in any performance analysis is...A 100%+ improvement in any performance analysis is a good result.<BR/><BR/>It's a shame we have to rely on standard tests such as DBT2 for comparison when these don't always represent the real world.<BR/><BR/>I would be good for the readers to also know the specifics of the SSD in use, as there is some variance between providers.Ronald Bradfordhttps://www.blogger.com/profile/16170615042077930093noreply@blogger.comtag:blogger.com,1999:blog-24359421.post-37557363979726126312009-03-06T17:11:00.000+01:002009-03-06T17:11:00.000+01:00Out of curiousity, was InnoDB in your test set to ...Out of curiousity, was InnoDB in your test set to flush at 0,1, or 2?<BR/><BR/>Either way some very promising results!ryanhttps://www.blogger.com/profile/04325518011075898953noreply@blogger.comtag:blogger.com,1999:blog-24359421.post-67231742795101164532009-03-06T16:02:00.000+01:002009-03-06T16:02:00.000+01:00Well done, Paul!It seems to me that, by looking in...Well done, Paul!<BR/>It seems to me that, by looking into DBT2 performance, you have found and fixed some general purpose bottlenecks.<BR/><BR/>Looking forward to seeing the next enhancements!<BR/><BR/>GiuseppeGiuseppe Maxiahttps://www.blogger.com/profile/15801583338057324813noreply@blogger.comtag:blogger.com,1999:blog-24359421.post-30186111101349685912009-03-06T15:40:00.000+01:002009-03-06T15:40:00.000+01:00PBXT is already a little rockstar:-)How is the HA ...PBXT is already a little rockstar:-)<BR/>How is the HA going? Read the paper from the MySQL Conference 2008, so Im curious.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-24359421.post-86822182343977499372009-03-06T15:21:00.000+01:002009-03-06T15:21:00.000+01:00Paul, congrats! great results ;)Looking forward to...Paul, congrats! <BR/><BR/>great results ;)<BR/><BR/>Looking forward to hearing about parallel index updates at the UC.<BR/><BR/>Rolandrpboumanhttps://www.blogger.com/profile/13365137747952711328noreply@blogger.com