[Ace-users] [tao-users] scalability/thread pool performance (500+ clients)

Andrew Solodovnikov blackborn at yandex.ru
Sat Mar 1 07:39:11 CST 2008

DCS> Hi Andrew,

Douglas, thanks for answer.

>>This causes 0% server load with a single orb thread, and 50-60% with a
>>pool, with no perfomance gain at all.
>>OS - Windows XP/2003, Intel Core 2 duo, 100 mbit network.

DCS> I recommend you check out

DCS> TAO_ROOT/docs/configurations.html

Yes, i saw it. And have tried most of them (MT server with thread
pool, ST server, ST server with select_st) and various factory options
- nope. It doesn't scale anywhere. 150-200 connections - and no more
:( No difference between ST and MT models - moreover, ST doesn't load
the CPU at all (0-1%), while MT with thread pool brings load up to
50-60%. With no dispatch performance gain.

About thread-per-connection... I think, thread-per-connection option
is not a case - 500+ threads is a very big price for scalability.

DCS> and try using various concurrency models to see which one performs the
DCS> best for your usecase.  For example, try using "multiple threads,
DCS> single ORB, thread-per-connection" and see how that compares with
DCS> "multiple threads, single ORB, TAO thread-pool model".  Both
DCS> approaches should work better than the default reactive model.

DCS> Take care,

DCS>      Doug


Andrew Solodovnikov

More information about the Ace-users mailing list