Hi Marco, I'll try to write some application reproducing this behavior this weekend. Maybe Monday i'll send it to this list. The application that i have is impossible, it has about 50 classes and its code has a few thousands of lines.
Manuel ----- Original Message ----- From: "Marco Russo" <[EMAIL PROTECTED]> To: <[EMAIL PROTECTED]> Sent: Friday, May 31, 2002 4:54 PM Subject: Re: [ADVANCED-DOTNET] singleton server problem > Manuel, > > I've no solution, but this is very different from what I observed with > remoting. My server stop to answer client request only if it goes to > 100% cpu processing a client request. With an Idle system, I never > observed a similar situation. > It would be interesting a sample to reproduce the same effect without > all your complex simulation system... Just to make some test... > > Marco > > -----Original Message----- > From: Moderated discussion of advanced .NET topics. > [mailto:[EMAIL PROTECTED]] On Behalf Of Manuel Costa > Sent: venerd́ 31 maggio 2002 15.17 > To: [EMAIL PROTECTED] > Subject: Re: [ADVANCED-DOTNET] singleton server problem > > > I will expose my situation in detail otherwise i think most people will > not understand what is happening. I'm using remoting as a tool for > communications in the distributed simulation of a pollutant transport in > a water body, like an estuary. Parallelization is achieved by domain > decomposition, assigning a peace of the 2D computational domain to a > machine. One machine, with one processor can receive several partitions > (small spatial regions) in order to mitigate the possibility of load > imbalances. Therefore, for each machine i have one thread per partition, > allowing only one to run. To do so i use a Mutex. To minimize the > communication cost i have to find which region inside the partition is > influencing the neighboring regions. After that i have to apply the > numerical method to them and asynchronously send pollutant mass to the > neighbors and at the same time apply the numerical method to those > having no influence on neighbors. Thereby, per machine i will have a > main thread responsible for the simulation and those as a consequence of > the asynchronous neighboring communications. The simulation thread runs > with a priority below normal. If for example, i have a cluster with 2 > machines and each machine with 9 partitions there is no problem. If i > increase the number of partitions per machine i will increase the number > of possible neighbors and therefore the number of secondary threads and > the problem relies here. After some number of secondary threads, the > remoting mechanism will completely stop answering to any new call. > Everything will stay stopped and the CPUs will stay idle. To test if it > was a problem with synchronization of my responsibility i have a button > on my end user application doing some stupid call to one of the server > components that i'm sure it has no problems with synchronization. The > call will never arrive to the server. The client will stay blocked and > my breakpoint at the server side will not capture anything. By all this, > i think there is a problem with some remoting class with > synchronization. An idle system not answering to nothing it really looks > like a synchronization issue. If anyone has some suggestion how to solve > this i'd really like to know it. > > Regards, > > Manuel > > You can read messages from the Advanced DOTNET archive, unsubscribe from > Advanced DOTNET, or subscribe to other DevelopMentor lists at > http://discuss.develop.com. > > You can read messages from the Advanced DOTNET archive, unsubscribe from Advanced DOTNET, or > subscribe to other DevelopMentor lists at http://discuss.develop.com. You can read messages from the Advanced DOTNET archive, unsubscribe from Advanced DOTNET, or subscribe to other DevelopMentor lists at http://discuss.develop.com.
