I'm using apache httpd 2.4.16 and attempting to do some network
benchmarking on windows, and noted extremely poor performance with
static files and large transfers through ajp to tomcat.
I decided on a lark to tune SendBufferSize to 131072 and bingo.
massive transfer speed difference.
Without SendBufferSize set, I'm pushing <20MB/s or so. This is on a
local gigabit LAN.
Once I set SendBufferSize to 131072 my transfers increased to about 30MB/s.
Tuning the AJP packet size resulted in an even larger increase (going
64k got me to 60MB/s - which is pretty close to what I would expect
given other forms of network file transfers.
Tomcat and apache are both local, so I really don't get why the packet
size made such a big difference but what I especially don't understand
is why the SendBufferSize actually matters. I was under the impression
that with older versions of windows, pre XP-SP2, the SendBufferSize
defaulted to 8k (which actually explains the <20MB/s performance) but
modern windows I thought they'd fixed that? Is this not the case? Is
the default sendbuffer still only 8192bytes?
Thanks,
Andy
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@httpd.apache.org
For additional commands, e-mail: users-h...@httpd.apache.org
