Emmanuel, are the Linux tests run on a VM or on a native Linux box ?
Jeff On Tue, May 28, 2013 at 4:21 PM, Emmanuel Lécharny <[email protected]>wrote: > Sorry, one slight mistake I tested with 1M messages, not 1G ! > > Le 5/28/13 3:55 PM, Emmanuel Lécharny a écrit : > > Hi guys, > > > > as Juien has coded the UDP client, I was able to run the benchmarks, so > > here are the results (dne on an old linux ox with a Java 7 jdk) > > > > All the tests are done with 1G messages being sent, except the last one, > > which sends 500K messages > > > > BIO vs BIO : > > ------------ > > 10 bytes 42.653s > > 1024 bytes 41.730s > > 2048 bytes 42.718s > > 4096 bytes 44.807s > > 8192 bytes 26.087s > > > > This is a pure BIO test, used as a reference > > > > NIO vs NIO, Mina : > > ------------------ > > > > 10 bytes 65.326s > > 1024 bytes 65.922s > > 2048 bytes 67.312s > > 4096 bytes 72.390s > > 8192 bytes 41.318s > > > > NIO vs NIO, Netty 3 : > > --------------------- > > > > 10 bytes 70.825s > > 1020 bytes 73.174s (but I had to remove 4 bytes) > > 2048 bytes N/A > > 4096 bytes N/A > > 8192 bytes N/A > > > > Some comments : > > > > - on my MacOSX box, the NIO tests are way faster : they are almost on > > par with BIO > > - all those tests have been ran on one single machine > > - Netty3 does not allow you to send more than 1024 bytes, max (this is > > why I'm sending 1020 bytes in the test, because 4 bytes are used for the > > data length) > > > > Bottom line, we are 50% slower than with BIO, and around 10% faster than > > Netty3. > > > > I haven't yet tested with Netty4, as the API have been completely > > rewritten, for the pain of the users :/ > > Working on it. > > > > > -- > Regards, > Cordialement, > Emmanuel Lécharny > www.iktek.com > > -- Jeff MAURY "Legacy code" often differs from its suggested alternative by actually working and scaling. - Bjarne Stroustrup http://www.jeffmaury.com http://riadiscuss.jeffmaury.com http://www.twitter.com/jeffmaury
