Remember to disable CPU frequency scaling when benchmarking... some things with IO cause the freq to drop, and when it's CPU bound again it takes a while for Linux to scale up the freq again.
For example, on my ubuntu box, ChannelFile went from 100MB/sec to 388MB/sec. This effect probably won't be uniform across implementations either. -Yonik http://www.lucidimagination.com On Tue, Sep 15, 2009 at 3:25 PM, Mark Miller <markrmil...@gmail.com> wrote: > I just really I hadn't sent this one. Here are results from the harddrive: > > It looks like its closer to the same speed on the hardrive once > everything is loaded in the system cache (as you'd expect). SeparateFile > was 1200 vs almost 1700 on RAMDISK. ChannelPread looked a lot closer though. > > > - Mark > > Tests from harddisk (filesystem cache warmed): > > config: impl=SeparateFile serial=false nThreads=4 iterations=100 > bufsize=1024 poolsize=2 filelen=730554368 > answer=-282293977, ms=238230, MB/sec=1226.6370616630988 > > config: impl=ChannelPread serial=false nThreads=4 iterations=100 > bufsize=1024 poolsize=2 filelen=730554368 > answer=-282295361, ms=766340, MB/sec=381.3212767179059 > > > Mark Miller wrote: >> Michael McCandless wrote: >> >>> I don't like that the answer is different... but it's really really >>> odd that it's different-yet-almost-the-same. >>> >>> Mark, were these 4 results on a normal (ext4) filesystem, or tmpfs? >>> (Because the top 2 entries of your 4 results match the first set of 2 >>> entries you sent... so I'm thinking these 4 were actually tmpfs not >>> ext4). >>> >>> >> Those 4 were tmpfs - I mention ext4 at the end because I had just given >> a feel for the hardrive tests and wanted to note it was from ext4 - the >> results are def ramdisk though. >> >>> What JRE/OS, linux, kernel versions, and hardware, are you running on? >>> >>> >> These are on: >> Ubuntu Karmic Koala 9.10, currently updated >> java-1.5.0-sun-1.5.0.20 >> 2.6.31-10-generic >> >> RAM: 3.9 Gig >> 4 core Intel Core2 duo 2.0GHz >> >> Slow 5200 rpm laptop drives. >> >> >>> The gains of SeparateFile over all else are stunning. And, quite >>> different from the linux tests I had run under LUCENE-753. Maybe we >>> need to revert FSDir.open to return SimpleFSDir again, on non-Windows >>> hosts. But then we don't have good concurrency... >>> >>> Mike >>> >>> On Tue, Sep 15, 2009 at 2:59 PM, Yonik Seeley >>> <yonik.see...@lucidimagination.com> wrote: >>> >>> >>>> It's been a while since I wrote that benchmarker... is it OK that the >>>> answer is different? Did you use the same test file? >>>> >>>> -Yonik >>>> http://www.lucidimagination.com >>>> >>>> >>>> >>>> On Tue, Sep 15, 2009 at 2:18 PM, Mark Miller <markrmil...@gmail.com> wrote: >>>> >>>> >>>>> The results: >>>>> >>>>> config: impl=SeparateFile serial=false nThreads=4 iterations=100 >>>>> bufsize=1024 poolsize=2 filelen=730554368 >>>>> answer=-282295611, ms=173550, MB/sec=1683.7899579371938 >>>>> >>>>> config: impl=ChannelFile serial=false nThreads=4 iterations=100 >>>>> bufsize=1024 poolsize=2 filelen=730554368 >>>>> answer=-282295361, ms=1377768, MB/sec=212.09793463050383 >>>>> >>>>> config: impl=ChannelPread serial=false nThreads=4 iterations=100 >>>>> bufsize=1024 poolsize=2 filelen=730554368 >>>>> answer=-282295361, ms=632253, MB/sec=462.19115955163517 >>>>> >>>>> config: impl=PooledPread serial=false nThreads=4 iterations=100 >>>>> bufsize=1024 poolsize=2 filelen=730554368 >>>>> answer=-282295361, ms=774664, MB/sec=377.2238637654518 >>>>> >>>>> ClassicFile was heading for the same fate as ChannelFile. >>>>> >>>>> >>>>> I'll have to check what its like on the file system - but it appears >>>>> just ridiculously slower. Even with SeparateFile, All 4 cores are bouncing >>>>> from 0-12% independently and really favoring the low end of that. >>>>> ChannelPread appears no better. >>>>> >>>>> There are results from other OS's/setups in the JIRA issue. >>>>> >>>>> I'm using ext4. >>>>> >>>>> Uwe Schindler wrote: >>>>> >>>>> >>>>>> How does a conventional file system compare? >>>>>> >>>>>> ----- >>>>>> Uwe Schindler >>>>>> H.-H.-Meier-Allee 63, D-28213 Bremen >>>>>> http://www.thetaphi.de >>>>>> eMail: u...@thetaphi.de >>>>>> >>>>>> >>>>>> >>>>>> >>>>>>> -----Original Message----- >>>>>>> From: Mark Miller [mailto:markrmil...@gmail.com] >>>>>>> Sent: Tuesday, September 15, 2009 7:15 PM >>>>>>> To: java-user@lucene.apache.org >>>>>>> Subject: Re: lucene 2.9.0RC4 slower than 2.4.1? >>>>>>> >>>>>>> Mark Miller wrote: >>>>>>> >>>>>>> >>>>>>> >>>>>>>> Indeed - I just ran the FileReaderTest on a Linux tmpfs ramdisk - with >>>>>>>> SeparateFile all 4 of my cores are immediately pinned and remain so. >>>>>>>> With ChannelFile, all 4 cores hover 20-30%. >>>>>>>> >>>>>>>> It would appear it may not be a good idea to use NIOFSDirectory on >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>> ramdisks. >>>>>>> >>>>>>> >>>>>>> >>>>>>>> Even still though - it looks like you have a further issue - your >>>>>>>> Lucene >>>>>>>> 2.9 old-api results don't use it, and are still not good. >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>> The quick results: >>>>>>> >>>>>>> ramdisk: sudo mount -t tmpfs tmpfs /tmp/space -o >>>>>>> size=1G,nr_inodes=200k,mode=01777 >>>>>>> >>>>>>> config: impl=SeparateFile serial=false nThreads=4 iterations=100 >>>>>>> bufsize=1024 poolsize=2 filelen=730554368 >>>>>>> answer=-282295611, ms=173550, MB/sec=1683.7899579371938 >>>>>>> >>>>>>> config: impl=ChannelFile serial=false nThreads=4 iterations=100 >>>>>>> bufsize=1024 poolsize=2 filelen=730554368 >>>>>>> answer=-282295361, ms=1377768, MB/sec=212.09793463050383 >>>>>>> >>>>>>> >>>>>>> -- >>>>>>> - Mark >>>>>>> >>>>>>> http://www.lucidimagination.com >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> --------------------------------------------------------------------- >>>>>>> To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org >>>>>>> For additional commands, e-mail: java-user-h...@lucene.apache.org >>>>>>> >>>>>>> >>>>>>> >>>>>> --------------------------------------------------------------------- >>>>>> To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org >>>>>> For additional commands, e-mail: java-user-h...@lucene.apache.org >>>>>> >>>>>> >>>>>> >>>>>> >>>>> -- >>>>> - Mark >>>>> >>>>> http://www.lucidimagination.com >>>>> >>>>> >>>>> >>>>> >>>>> --------------------------------------------------------------------- >>>>> To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org >>>>> For additional commands, e-mail: java-user-h...@lucene.apache.org >>>>> >>>>> >>>>> >>>>> >>>> --------------------------------------------------------------------- >>>> To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org >>>> For additional commands, e-mail: java-user-h...@lucene.apache.org >>>> >>>> >>>> >>>> >>> --------------------------------------------------------------------- >>> To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org >>> For additional commands, e-mail: java-user-h...@lucene.apache.org >>> >>> >>> >> >> >> > > > -- > - Mark > > http://www.lucidimagination.com > > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org > For additional commands, e-mail: java-user-h...@lucene.apache.org > > --------------------------------------------------------------------- To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org For additional commands, e-mail: java-user-h...@lucene.apache.org