I tried to modify FileManager to just use a BufferedOutputStream but discovered I couldn’t as the layouts now require the ByteBuffer.
Ralph > On Feb 8, 2017, at 12:14 AM, Apache <ralph.go...@dslextreme.com> wrote: > > The append method isn’t synchronized but the writeBytes method acquires a > lock. His code is actually a lot simpler than ours in that it just uses a > BufferedOutputStream and he only obtains the lock when he is writing to it. > > Ralph > >> On Feb 6, 2017, at 5:23 PM, Matt Sicker <boa...@gmail.com >> <mailto:boa...@gmail.com>> wrote: >> >> I'm not sure if I'm looking in the right place, but a major difference now >> between Logback's appenders and Log4j's is that Logback isn't synchronized >> on the append method. >> >> On 6 February 2017 at 18:18, Matt Sicker <boa...@gmail.com >> <mailto:boa...@gmail.com>> wrote: >> Is this something we can improve performance on by implementing a file >> appender based on FileChannel or AsynchronousFileChannel instead of >> OutputStream? >> >> On 6 February 2017 at 17:50, Apache <ralph.go...@dslextreme.com >> <mailto:ralph.go...@dslextreme.com>> wrote: >> Ceki has updated his numbers to include those reported on the mailing list. >> https://docs.google.com/spreadsheets/d/1cpb5D7qnyye4W0RTlHUnXedYK98catNZytYIu5D91m0/edit#gid=0 >> >> <https://docs.google.com/spreadsheets/d/1cpb5D7qnyye4W0RTlHUnXedYK98catNZytYIu5D91m0/edit#gid=0> >> >> I haven’t run the tests with Logback 1.2-SNAPSHOT but my numbers for my two >> MacBooks are at >> https://docs.google.com/spreadsheets/d/1L67IhmUVvyLBWtC6iq0TMj-j0vrbKsUKWuZV0Nlqisk/edit?usp=sharing >> >> <https://docs.google.com/spreadsheets/d/1L67IhmUVvyLBWtC6iq0TMj-j0vrbKsUKWuZV0Nlqisk/edit?usp=sharing>. >> >> >> Ralph >> >>> On Feb 6, 2017, at 9:33 AM, Apache <ralph.go...@dslextreme.com >>> <mailto:ralph.go...@dslextreme.com>> wrote: >>> >>> Yes, that is still the standard approach most people use and is the only >>> way to provide a head-to-head comparison against the logging frameworks. >>> >>> Ralph >>> >>>> On Feb 6, 2017, at 8:02 AM, Matt Sicker <boa...@gmail.com >>>> <mailto:boa...@gmail.com>> wrote: >>>> >>>> This is all in a synchronous appender, right? Either way, that's rather >>>> interesting. >>>> >>>> On 6 February 2017 at 07:54, Apache <ralph.go...@dslextreme.com >>>> <mailto:ralph.go...@dslextreme.com>> wrote: >>>> Someone posted numbers on the Logback user’s list that match mine. It >>>> shows Logback 1.1.9 was pretty terrible, 1.1.10 is somewhat better and >>>> 1.2-SNAPSHOT is on par or slightly better than Log4j 2. >>>> >>>> Ralph >>>> >>>>> On Feb 5, 2017, at 3:25 PM, Matt Sicker <boa...@gmail.com >>>>> <mailto:boa...@gmail.com>> wrote: >>>>> >>>>> I think we need some comparisons on the log4j side: file appender with >>>>> 256k buffer size, random access file appender with 256k buffer size >>>>> (which appears to be the default), and memory mapped file appender. It'd >>>>> be cool to see how these compose with async logging enabled in both log4j >>>>> and logback. >>>>> >>>>> On 5 February 2017 at 16:06, Apache <ralph.go...@dslextreme.com >>>>> <mailto:ralph.go...@dslextreme.com>> wrote: >>>>> You should run the code at https://github.com/ceki/logback-perf >>>>> <https://github.com/ceki/logback-perf> to compare your results to Ceki’s. >>>>> You also should capture the cpubenchmark speed of your processor and get >>>>> the speed of your hard drive. I used Blackmagic speed test on my Mac. I >>>>> am capturing my results in a Google spreadsheet. I will post the like >>>>> once I have it. >>>>> >>>>> Ralph >>>>> >>>>>> On Feb 5, 2017, at 1:48 PM, Gary Gregory <garydgreg...@gmail.com >>>>>> <mailto:garydgreg...@gmail.com>> wrote: >>>>>> >>>>>> If you want, I can run tests on Windows once the build works on Windows >>>>>> again. >>>>>> >>>>>> Let me know what args/command line... >>>>>> >>>>>> Gary >>>>>> >>>>>> On Feb 5, 2017 11:58 AM, "Apache" <ralph.go...@dslextreme.com >>>>>> <mailto:ralph.go...@dslextreme.com>> wrote: >>>>>> I guess my MacBook Pro must fit in the Slow CPU/Fast Hard drive >>>>>> category. With Logback 1.10 and -t 4 now get >>>>>> >>>>>> Benchmark Mode Samples >>>>>> Score Error Units >>>>>> o.a.l.l.p.j.FileAppenderBenchmark.julFile thrpt 20 >>>>>> 98187.673 ± 4935.712 ops/s >>>>>> o.a.l.l.p.j.FileAppenderBenchmark.log4j1File thrpt 20 >>>>>> 842374.496 ± 6762.712 ops/s >>>>>> o.a.l.l.p.j.FileAppenderBenchmark.log4j2File thrpt 20 >>>>>> 1853062.583 ± 67032.225 ops/s >>>>>> o.a.l.l.p.j.FileAppenderBenchmark.log4j2RAF thrpt 20 >>>>>> 2036011.226 ± 53208.281 ops/s >>>>>> o.a.l.l.p.j.FileAppenderBenchmark.logbackFile thrpt 20 >>>>>> 999667.438 ± 12074.003 ops/s >>>>>> >>>>>> I’ll have to try this on one my VMs at work. We don’t run anything >>>>>> directly on bare metal any more. >>>>>> >>>>>> Ralph >>>>>> >>>>>>> On Feb 5, 2017, at 9:40 AM, Apache <ralph.go...@dslextreme.com >>>>>>> <mailto:ralph.go...@dslextreme.com>> wrote: >>>>>>> >>>>>>> Ceki finally fixed some of the performance problems in the >>>>>>> FileAppender. See https://logback.qos.ch/news.html >>>>>>> <https://logback.qos.ch/news.html> and >>>>>>> https://docs.google.com/spreadsheets/d/1cpb5D7qnyye4W0RTlHUnXedYK98catNZytYIu5D91m0/edit#gid=0 >>>>>>> >>>>>>> <https://docs.google.com/spreadsheets/d/1cpb5D7qnyye4W0RTlHUnXedYK98catNZytYIu5D91m0/edit#gid=0>. >>>>>>> I suspect we have a few optimizations we can make. >>>>>>> >>>>>>> Ralph >>>>>> >>>>> >>>>> >>>>> >>>>> >>>>> -- >>>>> Matt Sicker <boa...@gmail.com <mailto:boa...@gmail.com>> >>>> >>>> >>>> >>>> >>>> -- >>>> Matt Sicker <boa...@gmail.com <mailto:boa...@gmail.com>> >>> >> >> >> >> >> -- >> Matt Sicker <boa...@gmail.com <mailto:boa...@gmail.com>> >> >> >> >> -- >> Matt Sicker <boa...@gmail.com <mailto:boa...@gmail.com>> >
