Hi! On 03:18 Tue 12 Aug , Mohammed Mustaq wrote: > Hi . > > ???I am not sured whether this is correct mailing List to post this message ? Any help would appreciate .? > > ????I have more more then 30 threads in a single process . I want to have sanity control for all threads i.e suppose either some thread has been stopped or some have been locked ( may be dead lock ) . > ???This could be done by having message Qs in all threads and broadcasting messages by any single thread?to?all and getting some reply .Except this , > > ??Is there any mechanism that thread manager can watch all threads ? or Any Other ? > > ??Am i missing something ? I think that is a very bad idea. Threading synchronisation is *hard* and deadlocks are just a part. If you do not know when threads stop, a "thread manager" will not help you. You can try these things: - Avoid using threads: many io operations can be done in a non-blocking way and multiplexed with e.g. epoll. Timers can be also handled by this thread. This is often faster than creating multiple threads in is quite often done (e.g. lighttpd) - Do not use locked shared memory, but message passing and/or transaction based "shared memory" (e.g. database, software emulated transactional memory) to do the synchronisation. Be aware of what happens when message passing queues get big. If they grow with no limit, your application might eat easily "eat" all available memory. If there is a limit, you might run into a deadlock, when 2 threads wait until they can pass a message to the other one. - Give the thread synchronisation the attention it deserves. It is not something you can can do on trial-and-error, because in many cases, you will not see the error. It will just appear on some machines depending on number an type of cpu, the process scheduler, amount of memory (swapping), other programs running in the background. This is why many people who know what they are doing *hate* threading and often try to avoid it. Actually it introduces a completly now type of bugs. Trying to improve performance without, any *measurement* can be called "premature optimisation" and this is a rather bad idea. -Michi -- programing a layer 3+4 network protocol for mesh networks see http://michaelblizek.twilightparadox.com -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
