On Tue, 02 Jun 2009 08:11:52 +0200, Kristian Nielsen <[email protected]> wrote: > [email protected] writes: > >> Benchmarking did not show any significant benefit of having parallel >> applying. >> Reason is that RBR applying is so fast compared to processing >> of SQL queries that applying queue did not build up. It turned out that >> several applier threads were useful only when transaction length was ~200 >> SQL >> statements. > > Sounds like you are testing a CPU bound load with all data in memory? > > If you test against a disk-bound load with data spread out over multiple > disk > drives, you would probably see a much higher benefit. A single-threaded > replication thread is not able to utilise the possibility to have multiple > outstanding I/O running in parallel on multiple disks. > > - Kristian. >
I wonder how practicable such setup really is. If one has disk-bound application, would not he rather use RAID or something to speed up individual writes than trying to execute writes in parallel hoping that they get to different drives? I mean acceleration of writes on low level always works, parallelization would require suitable application profile and careful distribution of the database on multiple drives. Or are we talking about multiple RAIDs? I think we can expect a benefit from parallel applying exactly on CPU-bound load on a 4-core or higher machines where one core just won't be able to keep up. But then again, optimizing binlog events (by, say, making them storage specific, or optimizing storage engine to use several drives) may bring a greater and earlier benefit considering what a pain parallel transaction applying promises to be. And pardon my ignorance, are there any figures that substantiate the necessity of parallel applying (I mean measured gains) or is it just a speculation that parallel applying is a good thing to have? Alex _______________________________________________ Mailing list: https://launchpad.net/~drizzle-discuss Post to : [email protected] Unsubscribe : https://launchpad.net/~drizzle-discuss More help : https://help.launchpad.net/ListHelp
