Hi, I guess all the results here are for local files. We log to network file shares over SMB. So I decided to run the PerformanceComparison test from this thread on Windows XP and Java 1.6.0_07 for local and remote files. Here are the results... ############################################### local, FileOutputStream: 7031 local, BufferedOutputStream: 3008 local, FileWriter: 2653 local, FileChannel: 8259 ############################################### ############################################### remote, FileOutputStream: 1618729 remote, BufferedOutputStream: 8743 remote, FileWriter: 8191 remote, FileChannel: 1623684 ############################################### So can we make logback use FileWriter? It seems to be the best for both local and remote files. Also, bizarely, after the test had ran, when I look in the local and remote folders that I was writting to, there were the testbuffer.log and testos.log files, but there was no testchannel.log file? I'm not NIO expert, and the FileOutputStream is being closed, so what's going wrong? David Peter-262 wrote:
Yeah, with few more (and hopefuly last) things to add:
Not flushing after every line doesn't seem to make a difference for FileOutputStream in isolated tests (I was getting identical times as with flushing) - unless buffering of course - but it does make difference with logback which uses this FileOutputStream. However logback wraps OutputStreamWriter around it and then it really makes that difference - and also results become very close (almost identical) to FileWriter then.
Java 1.5 only: OutputStreamWriter/FileOutputStream immediateFlush: 8435 OutputStreamWriter/FileOutputStream no-flush: 1160 BufferedWriter/OutputStreamWriter/FileOutputStream no-flush: 922
FileWriter immediateFlush: 8271 FileWriter no-flush: 1150 BufferedWriter/FileWriter no-flush: 918
In real logging scenarios it looks to me that immediateFlush=false decreases times to 1/2-1/3 and buffering can give further 10-15% improvement with properly sized buffer.
So I definitely vote for bringing at least the immediateFlush=false option back. Bringing back buffering would also be great as one can get a bit more performance out of it - and every little bit does matter for us here. :) _______________________________________________ Logback-user mailing list Logback-user@qos.ch http://qos.ch/mailman/listinfo/logback-user
-- View this message in context: http://old.nabble.com/log4j-performance-comparison-tp31183677p31368496.html Sent from the Logback User mailing list archive at Nabble.com.