Hi Romain,

I use RMIIO extensively for large binary transfer that I modified with an
almost identical fix to your flushable gzip implementation. I have tested
this to death with and without buffering/gzip in the past.

I have found that gzip works substantially better with buffered streams over
a 'real' network, testing locally is unlikely to reveal massive gains on
small data - The gzip streams have a 4k buffer, buffered streams an 8k.

The difference in IO code between OpenEJB and RMIIO for this sake is
trivial, so I have made an assumption (educated guess if you like) that this
can only be of benefit.

One RMIIO cycling 50mb random binary write test I have is along the lines
of:

client - buffer - gzip - network - buffer - gunzip - server - blob = 5sec
av.

client - buffer - network - buffer - server - blob = 7sec av.

And just tested without buffering...
client - gzip - network - gunzip - server - blob = 12sec av.

So over a day quite a lot!

Andy.



--
View this message in context: 
http://openejb.979440.n4.nabble.com/Fwd-svn-commit-r1415122-in-openejb-trunk-openejb-server-openejb-ejbd-src-main-java-org-apache-openeja-tp4658985p4658989.html
Sent from the OpenEJB Dev mailing list archive at Nabble.com.

Reply via email to