Re: [OMPI devel] Need help for semaphore in BML
On Apr 19, 2007, at 1:45 PM, po...@cc.gatech.edu wrote: I want to put semaphore in bml.h--- mca_bml_send before and after calling btl_send. SO that when a process call btl_send it first lock a global variable X and then proceeds.Also if an external Tcp function wants to send data it should first lock global variable X and then proceed. Can anyone tell me only changing bml.h is enough or are there any other files where I need to make changes. This is likely to be a complex issue because there's the put and get functions as well. ob1 uses a fairly complex algorithm to decide when to call the bml interface functions -- I doubt that the use of a semaphore in a single location is going to do what you want. (why a semaphore, anyway -- why not a mutex?) (As I tried doing this and run mpi program it gave me ORTE time out error also when I changed file back to normal it was not compiling and giving me error in libmca_bml.la etc...unfortunately I deleted entire folder and downloaded new version.) Changing bml.h should have zero effect on the ORTE layer. ORTE is a whole different abstraction and wholly below the OMPI layer. There are a few places in the OMPI layer that interact with the lower ORTE later, but the bml is not one of them. I'm guessing that you had some other problem. If you're going to be working continually with Open MPI, you might want to get a subversion checkout. Can any one please help me and tell me how should I go about implementing locks/semaphore in bml layer so that all mpi process access lock (of same priority ) and continue working while Tcp acquire only when network is free(or there is lot of serial operation between 2 mpi sends). I want to emphasize again that this won't give you what you have described in previous mails: the PML interface is designed to be asynchronous. So when you call send/put/get, it only (possibly) *starts* the communication transfer. When you unlock upon return, you're allowing the alternate communication mechanism to come in and start another communication method (via a different BTL, perhaps), but it does not change that there may still be activity occurring down in the kernel and/or hardware. Also, this scheme does not account for received message contention -- it only [tries to] account for sending contention. So even if you get the locking working the way that you want, I don't think that you're going to get the overlap and multiplexing that you expect. -- Jeff Squyres Cisco Systems
Re: [OMPI devel] SOS... help needed :(
Are both the IB HCA and the ethernet interfaces on the same physical bus? If they're not, the need for multiplexing them is diminished (but, of course, it depends on what you're trying to do -- if everything is using huge memory transfers, then your bottleneck will be RAM, not the bus that the NICs reside on). That being said, something we have not explored at all is the idea of multiplexing at the MPI layer. Perhaps something like "this is a low priority communicator; I want you to only use the 'tcp' BTL on it" and "this is a high priority communicator; I want you to only use the 'openib' BTL on it". I haven't thought at all about whether that is possible. It would probably take some mucking around in both the bml and the ob1 pml. Hmm. It may or may not be worth it, but I raise the possibility... On Apr 19, 2007, at 9:18 PM, po...@cc.gatech.edu wrote: Hi, Some of our clusters uses Gigabit Ethernet and Infiniband. So we are trying to multiplex them. Thanks and Regards Pooja On Thu, Apr 19, 2007 at 06:58:37PM -0400, po...@cc.gatech.edu wrote: I am Pooja working with chaitali on this project. The idea behind this is while running a parallelized code ,if a huge chunks of serial computation is encountered at that time underlying network infrastructure can be used for some other data transfer. This increases the network utilization. But this (non Mpi) data transfer should not keep Mpi calls blocking. So we need to give them priorities. Also we are trying to predict a behavior of the code (like if there are more MPi calls coming with short interval or if they are coming after large interval ) based on previous calls. As a result we can make this mechanism more efficient. Ok, so you have a Cluster with Infiniband a while the network traffic is low you want to utilize the Infiniband network for other data transfers with a lower priority? What does this have to do with TCP or are you using TCP over Infiniband? Regards Christian Leber -- http://rettetdieti.vde-uni-mannheim.de/ ___ devel mailing list de...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/devel ___ devel mailing list de...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/devel -- Jeff Squyres Cisco Systems
Re: [OMPI devel] SOS... help needed :(
Hi, Some of our clusters uses Gigabit Ethernet and Infiniband. So we are trying to multiplex them. Thanks and Regards Pooja > On Thu, Apr 19, 2007 at 06:58:37PM -0400, po...@cc.gatech.edu wrote: > >> I am Pooja working with chaitali on this project. >> The idea behind this is while running a parallelized code ,if a huge >> chunks of serial computation is encountered at that time underlying >> network infrastructure can be used for some other data transfer. >> This increases the network utilization. >> But this (non Mpi) data transfer should not keep Mpi calls blocking. >> So we need to give them priorities. >> Also we are trying to predict a behavior of the code (like if there are >> more MPi calls coming with short interval or if they are coming after >> large interval ) based on previous calls. >> As a result we can make this mechanism more efficient. > > Ok, so you have a Cluster with Infiniband a while the network traffic is > low you want to utilize the Infiniband network for other data transfers > with a lower priority? > > What does this have to do with TCP or are you using TCP over Infiniband? > > Regards > Christian Leber > > -- > http://rettetdieti.vde-uni-mannheim.de/ > > ___ > devel mailing list > de...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/devel >
Re: [OMPI devel] SOS... help needed :(
On Thu, Apr 19, 2007 at 06:58:37PM -0400, po...@cc.gatech.edu wrote: > I am Pooja working with chaitali on this project. > The idea behind this is while running a parallelized code ,if a huge > chunks of serial computation is encountered at that time underlying > network infrastructure can be used for some other data transfer. > This increases the network utilization. > But this (non Mpi) data transfer should not keep Mpi calls blocking. > So we need to give them priorities. > Also we are trying to predict a behavior of the code (like if there are > more MPi calls coming with short interval or if they are coming after > large interval ) based on previous calls. > As a result we can make this mechanism more efficient. Ok, so you have a Cluster with Infiniband a while the network traffic is low you want to utilize the Infiniband network for other data transfers with a lower priority? What does this have to do with TCP or are you using TCP over Infiniband? Regards Christian Leber -- http://rettetdieti.vde-uni-mannheim.de/
