Summary: | Lock-ups in TAO 2.1.5 in simple multithreaded client-server benchmark | ||
---|---|---|---|
Product: | TAO | Reporter: | Sergey Onuchin <sonuchin> |
Component: | ORB | Assignee: | DOC Center Support List (internal) <tao-support> |
Status: | NEW --- | ||
Severity: | major | ||
Priority: | P3 | ||
Version: | 2.1.5 | ||
Hardware: | x86 | ||
OS: | Linux | ||
Attachments: | MT_Client/run_test.pl replacement |
Created attachment 1436 [details] MT_Client/run_test.pl replacement I've reproduced the issue on 6.1.5 using MT_Client test from TAO package. Please take attached mytest.pl file (default run_test.pl is not passing proper config to the client - consider to fix that). Modify MT_Client/client.conf to contain: #static Client_Strategy_Factory "-ORBClientConnectionHandler RW" static Client_Strategy_Factory "-ORBClientConnectionHandler MT" static Client_Strategy_Factory "-ORBTransportMuxStrategy EXCLUSIVE" On my 4-core box ./mytest.pl gets locked almost 100%. If you have more cores, try to increase the number of threads in .pl file. If you remove/comment static Client_Strategy_Factory "-ORBTransportMuxStrategy EXCLUSIVE" then no locking occurs. In 6.0.5 no locking happens no matter which TransportMuxStrategy is in use. The problem seems to be related to changes in Leader-Follower code since TAO 2.0.5.