Warmup Handshake Timeouts

[mrochon]

Hi Trustin, performance testing is going very well on our Netty server. We're
handling ~100k clients per cluster node, very nice.

One thing we have had to work around so far is masses of handshake
connections happening when the server is idle.

Basically when we get around 1k simultaneous connects and ssl handshakes the
server sometimes can not complete a handshake fast enough (around 0.5% of
the time).

Basically when the server has been idle for a while and we get a flood of
connects the JVM cannot spin up threads in the thread pool fast enough to
complete the handshake before the socket read timeout occurs.

I have increased the number of idle threads we keep around but I would not
like to scale in that manner, as it could take resources away from tasks
that are designed to eat up idle cpu time.

Basically I was looking for a way to increase the socket read timeout
(SO_TIMEOUT) on the child sockets to give myself a little more time for
server warmup (I think I can bring this down to < 0.05% with a very modest
increase in timeout).

Any chance we can get that in?
-- 
View this message in context: http://netty-forums-and-mailing-lists.685743.n2.nabble.com/Warmup-Handshake-Timeouts-tp5456858p5456858.html
Sent from the Netty User Group mailing list archive at Nabble.com.