April 2015 - discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] How to use non-blocking send/receive without calling MPI_Wait
by Lei Shi 03 Apr '15

03 Apr '15

Sorry. I just take a quick look of that slides this afternoon. Got the details now. So can I make a quick conclusion that there is no way to avoid MPI_Wait or MPI_test with non-blocking send/receive. Even I don't care I have received the message or not. On Fri, Apr 3, 2015 at 5:23 PM, Jeff Squyres (jsquyres) <jsquyres(a)cisco.com> wrote: > On Apr 3, 2015, at 6:09 PM, Balaji, Pavan <balaji(a)anl.gov> wrote: > > > > I didn't look through the video, but what you are saying is not true. > You can free the request in the application (this simply reduces the > reference count to that request). MPI will still maintain one reference > count to the request and will only free it when the data transfer completes. > > That's correct. The point of that blog entry is that you have to tell MPI > that you're done with a request somehow, otherwise MPI *has* to keep those > resources allocated. > > MPI_TEST*/MPI_WAIT* are the usual ways to do this. MPI_REQUEST_FREE is > also a valid method, but, as has been stated, isn't advisable. > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: > http://www.cisco.com/web/about/doing_business/legal/cri/ > > _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] How to use non-blocking send/receive without calling MPI_Wait
by Jeff Squyres (jsquyres) 03 Apr '15

03 Apr '15

On Apr 3, 2015, at 6:09 PM, Balaji, Pavan <balaji(a)anl.gov> wrote: > > I didn't look through the video, but what you are saying is not true. You can free the request in the application (this simply reduces the reference count to that request). MPI will still maintain one reference count to the request and will only free it when the data transfer completes. That's correct. The point of that blog entry is that you have to tell MPI that you're done with a request somehow, otherwise MPI *has* to keep those resources allocated. MPI_TEST*/MPI_WAIT* are the usual ways to do this. MPI_REQUEST_FREE is also a valid method, but, as has been stated, isn't advisable. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] How to use non-blocking send/receive without calling MPI_Wait
by Dave Goodell (dgoodell) 03 Apr '15

03 Apr '15

On Apr 3, 2015, at 4:55 PM, Jeff Hammond <jeff.science(a)gmail.com> wrote: > MPI_Wait does not guarantee that the data sent with MPI_Isend is received. It merely means you can reuse the send buffer. If you want to be sure it has been received, use MPI_Issend (synchronous send). MPI_Ssend or MPI_Issend completing only guarantees that the receiver has posted the matching receive, not that the data has been completely transferred to the receiver's buffer. -Dave _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] How to use non-blocking send/receive without calling MPI_Wait
by Balaji, Pavan 03 Apr '15

03 Apr '15

> On Apr 3, 2015, at 4:46 PM, Lei Shi <lshi(a)ku.edu> wrote: > I think according to Jeff's post http://blogs.cisco.com/performance/dont-leak-mpi_requests. If we don't call MPI_Wait or test, the request object will not be released even the data transfer completes. I didn't look through the video, but what you are saying is not true. You can free the request in the application (this simply reduces the reference count to that request). MPI will still maintain one reference count to the request and will only free it when the data transfer completes. -- Pavan -- Pavan Balaji ✉️ http://www.mcs.anl.gov/~balaji _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] How to use non-blocking send/receive without calling MPI_Wait
by Jeff Hammond 03 Apr '15

03 Apr '15

MPI_Wait does not guarantee that the data sent with MPI_Isend is received. It merely means you can reuse the send buffer. If you want to be sure it has been received, use MPI_Issend (synchronous send). I assume that you don't actually care if the data has been received, in which case you should not use synchronous send, since it may be slower than regular send, due to additional synchronization. Jeff On Fri, Apr 3, 2015 at 2:46 PM, Lei Shi <lshi(a)ku.edu> wrote: > Pavan, > > In my case, I don't care the data is correct or not. I know it sounds > crazy at first time, but there are some numerical schemes designed for this > situation. > > I think according to Jeff's post > http://blogs.cisco.com/performance/dont-leak-mpi_requests. If we don't > call MPI_Wait or test, the request object will not be released even the > data transfer completes. > > I will test your suggestion to call MPI_Request_free to release > it manually. > > BTW, attached is my test result, > 1. The orange one uses MPI_ISend/MPI_IRecv with MPI_Wait to make sure the > data has been received successfully. The WTime is linear with iterations, > which is correct. > > 2. The blue one uses MPI_ISend and MPI_IRecv without calling MPI_Wait. > There are some jumps in WTime. Does it due to some resource leaking? I will > try to call MPI_Request_free to solve those jumps. > > 3. The green one uses MPI_IBsend and MPI_IRecv (should I use MPI_Recv with > MPI_IBsend) without calling MPI_Wait. It is even slower than the case 1 > with MPI_Wait. > > > > > On Fri, Apr 3, 2015 at 1:29 PM, Balaji, Pavan <balaji(a)anl.gov> wrote: > >> >> No, the request object will get released fine when the data transfer >> completes. It is a correct program. What I meant is that, since the >> application never knows when the request completes, the user application >> cannot use the data in any meaningful way, IMO. >> >> -- Pavan >> >> > On Apr 3, 2015, at 12:13 PM, Lei Shi <lshi(a)ku.edu> wrote: >> > >> > Pavan, >> > >> > Thanks. You mean since I don't call MPI_Wait at all, the request object >> does not get a chance to be released by the MPI library? Right? I will give >> it a try. Thanks again! >> > >> > >> > On Fri, Apr 3, 2015 at 10:12 AM, Balaji, Pavan <balaji(a)anl.gov> wrote: >> > >> > You can free the request with MPI_Request_free if you don't want to >> wait on it. I have no idea how you'll write a correct program without >> waiting for receive completions at least, though. >> > >> > -- Pavan >> > >> > > On Apr 2, 2015, at 11:44 PM, Lei Shi <lshi(a)ku.edu> wrote: >> > > >> > > Hi Junchao, >> > > >> > > Thanks for your reply. For my case, I don't want to check the data >> have been received or not. So I don't want to call MPI_Test or any function >> to verify it. But my problem is like if I ignore calling the MPI_Wait, just >> call Isend/Irev, my program freezes for several sec and then continues to >> run. My guess is probably I messed up the MPI library internal buffer by >> doing this. >> > > >> > > >> > > On Thu, Apr 2, 2015 at 7:25 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> >> wrote: >> > > Does MPI_Test fit your needs? >> > > >> > > --Junchao Zhang >> > > >> > > On Thu, Apr 2, 2015 at 7:16 PM, Lei Shi <lshi(a)ku.edu> wrote: >> > > I want to use non-blocking send/rev MPI_Isend/MPI_Irev to do >> communication. But in my case, I don't really care what kind of data I get >> or it is ready to use or not. So I don't want to waste my time to do any >> synchronization by calling MPI_Wait or etc API. >> > > >> > > But when I avoid calling MPI_Wait, my program is freezed several secs >> after running some iterations (after multiple MPI_Isend/Irev callings), >> then continues. It takes even more time than the case with MPI_Wait. So my >> question is how to do a "true" non-blocking communication without waiting >> for the data ready or not. Thanks. >> > > >> > > >> > > _______________________________________________ >> > > discuss mailing list discuss(a)mpich.org >> > > To manage subscription options or unsubscribe: >> > > https://lists.mpich.org/mailman/listinfo/discuss >> > > >> > > >> > > _______________________________________________ >> > > discuss mailing list discuss(a)mpich.org >> > > To manage subscription options or unsubscribe: >> > > https://lists.mpich.org/mailman/listinfo/discuss >> > > >> > > _______________________________________________ >> > > discuss mailing list discuss(a)mpich.org >> > > To manage subscription options or unsubscribe: >> > > https://lists.mpich.org/mailman/listinfo/discuss >> > >> > -- >> > Pavan Balaji ✉️ >> > http://www.mcs.anl.gov/~balaji >> > >> > _______________________________________________ >> > discuss mailing list discuss(a)mpich.org >> > To manage subscription options or unsubscribe: >> > https://lists.mpich.org/mailman/listinfo/discuss >> > >> >> -- >> Pavan Balaji ✉️ >> http://www.mcs.anl.gov/~balaji >> >> > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss > -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/ _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] How to use non-blocking send/receive without calling MPI_Wait
by Lei Shi 03 Apr '15

03 Apr '15

Pavan, In my case, I don't care the data is correct or not. I know it sounds crazy at first time, but there are some numerical schemes designed for this situation. I think according to Jeff's post http://blogs.cisco.com/performance/dont-leak-mpi_requests. If we don't call MPI_Wait or test, the request object will not be released even the data transfer completes. I will test your suggestion to call MPI_Request_free to release it manually. BTW, attached is my test result, 1. The orange one uses MPI_ISend/MPI_IRecv with MPI_Wait to make sure the data has been received successfully. The WTime is linear with iterations, which is correct. 2. The blue one uses MPI_ISend and MPI_IRecv without calling MPI_Wait. There are some jumps in WTime. Does it due to some resource leaking? I will try to call MPI_Request_free to solve those jumps. 3. The green one uses MPI_IBsend and MPI_IRecv (should I use MPI_Recv with MPI_IBsend) without calling MPI_Wait. It is even slower than the case 1 with MPI_Wait. On Fri, Apr 3, 2015 at 1:29 PM, Balaji, Pavan <balaji(a)anl.gov> wrote: > > No, the request object will get released fine when the data transfer > completes. It is a correct program. What I meant is that, since the > application never knows when the request completes, the user application > cannot use the data in any meaningful way, IMO. > > -- Pavan > > > On Apr 3, 2015, at 12:13 PM, Lei Shi <lshi(a)ku.edu> wrote: > > > > Pavan, > > > > Thanks. You mean since I don't call MPI_Wait at all, the request object > does not get a chance to be released by the MPI library? Right? I will give > it a try. Thanks again! > > > > > > On Fri, Apr 3, 2015 at 10:12 AM, Balaji, Pavan <balaji(a)anl.gov> wrote: > > > > You can free the request with MPI_Request_free if you don't want to wait > on it. I have no idea how you'll write a correct program without waiting > for receive completions at least, though. > > > > -- Pavan > > > > > On Apr 2, 2015, at 11:44 PM, Lei Shi <lshi(a)ku.edu> wrote: > > > > > > Hi Junchao, > > > > > > Thanks for your reply. For my case, I don't want to check the data > have been received or not. So I don't want to call MPI_Test or any function > to verify it. But my problem is like if I ignore calling the MPI_Wait, just > call Isend/Irev, my program freezes for several sec and then continues to > run. My guess is probably I messed up the MPI library internal buffer by > doing this. > > > > > > > > > On Thu, Apr 2, 2015 at 7:25 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> > wrote: > > > Does MPI_Test fit your needs? > > > > > > --Junchao Zhang > > > > > > On Thu, Apr 2, 2015 at 7:16 PM, Lei Shi <lshi(a)ku.edu> wrote: > > > I want to use non-blocking send/rev MPI_Isend/MPI_Irev to do > communication. But in my case, I don't really care what kind of data I get > or it is ready to use or not. So I don't want to waste my time to do any > synchronization by calling MPI_Wait or etc API. > > > > > > But when I avoid calling MPI_Wait, my program is freezed several secs > after running some iterations (after multiple MPI_Isend/Irev callings), > then continues. It takes even more time than the case with MPI_Wait. So my > question is how to do a "true" non-blocking communication without waiting > for the data ready or not. Thanks. > > > > > > > > > _______________________________________________ > > > discuss mailing list discuss(a)mpich.org > > > To manage subscription options or unsubscribe: > > > https://lists.mpich.org/mailman/listinfo/discuss > > > > > > > > > _______________________________________________ > > > discuss mailing list discuss(a)mpich.org > > > To manage subscription options or unsubscribe: > > > https://lists.mpich.org/mailman/listinfo/discuss > > > > > > _______________________________________________ > > > discuss mailing list discuss(a)mpich.org > > > To manage subscription options or unsubscribe: > > > https://lists.mpich.org/mailman/listinfo/discuss > > > > -- > > Pavan Balaji ✉️ > > http://www.mcs.anl.gov/~balaji > > > > _______________________________________________ > > discuss mailing list discuss(a)mpich.org > > To manage subscription options or unsubscribe: > > https://lists.mpich.org/mailman/listinfo/discuss > > > > -- > Pavan Balaji ✉️ > http://www.mcs.anl.gov/~balaji > > _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] How to use non-blocking send/receive without calling MPI_Wait
by Jeff Hammond 03 Apr '15

03 Apr '15

As far as I know, Ethernet is not good at making asynchronous progress in hardware the way e.g. InfiniBand is. I would have thought that a dedicated progress thread would help, but it seems you tried that. Did you use your own progress thread or MPICH_ASYNC_PROGRESS=1? Jeff On Fri, Apr 3, 2015 at 10:10 AM, Lei Shi <lshi(a)ku.edu> wrote: > Huiwei, > > Thanks for your email. Your answer leads to my another question about > asynchronous MPI communication. > > I'm trying to do an overlapped communication/computing to speedup my MPI > code. I read some papers comparing some different approaches to do the > overlapped communication. The "naive" overlapped communication > implementation, which only use non-blocking mpi Isend/Irecv and the hybrid > approach using OpenMP and MPI together. In the hybrid approach, a separated > thread is use to do all non-blocking communications. Just exactly as you > said, the results indicate that current MPI implementations do not > support true asynchronous communication. > > If I use the naive approach, my code with non-blocking or blocking > send/recv gives me almost the same performance in term of Wtime. All > communications are postponed to MPI_Wait. > > I have tried calling mpi_test to push library to do communication during > iterations. And try to use a dedicated thread to do communication and the > other thread to do computing only. However, the performance gains are very > small or no gain at all. I'm wondering it is due to the hardware. The > cluster I tested uses 10G Ethernet card. > > > Best, > > Lei Shi > > > > > > On Fri, Apr 3, 2015 at 8:49 AM, Huiwei Lu <huiweilu(a)mcs.anl.gov> wrote: > >> Hi Lei, >> >> As far as I know, all current MPI implementations do not support true >> asynchronous communication for now. i.e., If there is no MPI calls in your >> iterations, MPICH will not be able to make progress on communication. >> >> One solution is to poll the MPI runtime regularly to make progress by >> inserting MPI_Test to your iteration (even though you do not want to check >> the data). >> >> Another solution is to enable MPI's asynchronous progress thread to make >> progress for you. >> >> -- >> Huiwei >> >> On Thu, Apr 2, 2015 at 11:44 PM, Lei Shi <lshi(a)ku.edu> wrote: >> >>> Hi Junchao, >>> >>> Thanks for your reply. For my case, I don't want to check the data have >>> been received or not. So I don't want to call MPI_Test or any function to >>> verify it. But my problem is like if I ignore calling the MPI_Wait, just >>> call Isend/Irev, my program freezes for several sec and then continues to >>> run. My guess is probably I messed up the MPI library internal buffer by >>> doing this. >>> >>> On Thu, Apr 2, 2015 at 7:25 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> >>> wrote: >>> >>>> Does MPI_Test fit your needs? >>>> >>>> --Junchao Zhang >>>> >>>> On Thu, Apr 2, 2015 at 7:16 PM, Lei Shi <lshi(a)ku.edu> wrote: >>>> >>>>> I want to use non-blocking send/rev MPI_Isend/MPI_Irev to do >>>>> communication. But in my case, I don't really care what kind of data I get >>>>> or it is ready to use or not. So I don't want to waste my time to do any >>>>> synchronization by calling MPI_Wait or etc API. >>>>> >>>>> But when I avoid calling MPI_Wait, my program is freezed several secs >>>>> after running some iterations (after multiple MPI_Isend/Irev callings), >>>>> then continues. It takes even more time than the case with MPI_Wait. So my >>>>> question is how to do a "true" non-blocking communication without waiting >>>>> for the data ready or not. Thanks. >>>>> >>>>> _______________________________________________ >>>>> discuss mailing list discuss(a)mpich.org >>>>> To manage subscription options or unsubscribe: >>>>> https://lists.mpich.org/mailman/listinfo/discuss >>>>> >>>> >>>> >>>> _______________________________________________ >>>> discuss mailing list discuss(a)mpich.org >>>> To manage subscription options or unsubscribe: >>>> https://lists.mpich.org/mailman/listinfo/discuss >>>> >>> >>> >>> _______________________________________________ >>> discuss mailing list discuss(a)mpich.org >>> To manage subscription options or unsubscribe: >>> https://lists.mpich.org/mailman/listinfo/discuss >>> >> >> >> _______________________________________________ >> discuss mailing list discuss(a)mpich.org >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/discuss >> > > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss > -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/ _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] How to use non-blocking send/receive without calling MPI_Wait
by Balaji, Pavan 03 Apr '15

03 Apr '15

No, the request object will get released fine when the data transfer completes. It is a correct program. What I meant is that, since the application never knows when the request completes, the user application cannot use the data in any meaningful way, IMO. -- Pavan > On Apr 3, 2015, at 12:13 PM, Lei Shi <lshi(a)ku.edu> wrote: > > Pavan, > > Thanks. You mean since I don't call MPI_Wait at all, the request object does not get a chance to be released by the MPI library? Right? I will give it a try. Thanks again! > > > On Fri, Apr 3, 2015 at 10:12 AM, Balaji, Pavan <balaji(a)anl.gov> wrote: > > You can free the request with MPI_Request_free if you don't want to wait on it. I have no idea how you'll write a correct program without waiting for receive completions at least, though. > > -- Pavan > > > On Apr 2, 2015, at 11:44 PM, Lei Shi <lshi(a)ku.edu> wrote: > > > > Hi Junchao, > > > > Thanks for your reply. For my case, I don't want to check the data have been received or not. So I don't want to call MPI_Test or any function to verify it. But my problem is like if I ignore calling the MPI_Wait, just call Isend/Irev, my program freezes for several sec and then continues to run. My guess is probably I messed up the MPI library internal buffer by doing this. > > > > > > On Thu, Apr 2, 2015 at 7:25 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> wrote: > > Does MPI_Test fit your needs? > > > > --Junchao Zhang > > > > On Thu, Apr 2, 2015 at 7:16 PM, Lei Shi <lshi(a)ku.edu> wrote: > > I want to use non-blocking send/rev MPI_Isend/MPI_Irev to do communication. But in my case, I don't really care what kind of data I get or it is ready to use or not. So I don't want to waste my time to do any synchronization by calling MPI_Wait or etc API. > > > > But when I avoid calling MPI_Wait, my program is freezed several secs after running some iterations (after multiple MPI_Isend/Irev callings), then continues. It takes even more time than the case with MPI_Wait. So my question is how to do a "true" non-blocking communication without waiting for the data ready or not. Thanks. > > > > > > _______________________________________________ > > discuss mailing list discuss(a)mpich.org > > To manage subscription options or unsubscribe: > > https://lists.mpich.org/mailman/listinfo/discuss > > > > > > _______________________________________________ > > discuss mailing list discuss(a)mpich.org > > To manage subscription options or unsubscribe: > > https://lists.mpich.org/mailman/listinfo/discuss > > > > _______________________________________________ > > discuss mailing list discuss(a)mpich.org > > To manage subscription options or unsubscribe: > > https://lists.mpich.org/mailman/listinfo/discuss > > -- > Pavan Balaji ✉️ > http://www.mcs.anl.gov/~balaji > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss > -- Pavan Balaji ✉️ http://www.mcs.anl.gov/~balaji _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] How to use non-blocking send/receive without calling MPI_Wait
by Huiwei Lu 03 Apr '15

03 Apr '15

Hi Lei, I think there should be some difference between blocking send and nonblocking send, regardless of what hardware you used. How did you overlap the computation with communication? Can you give us a snippet of you code? -- Huiwei On Fri, Apr 3, 2015 at 12:10 PM, Lei Shi <lshi(a)ku.edu> wrote: > Huiwei, > > Thanks for your email. Your answer leads to my another question about > asynchronous MPI communication. > > I'm trying to do an overlapped communication/computing to speedup my MPI > code. I read some papers comparing some different approaches to do the > overlapped communication. The "naive" overlapped communication > implementation, which only use non-blocking mpi Isend/Irecv and the hybrid > approach using OpenMP and MPI together. In the hybrid approach, a separated > thread is use to do all non-blocking communications. Just exactly as you > said, the results indicate that current MPI implementations do not > support true asynchronous communication. > > If I use the naive approach, my code with non-blocking or blocking > send/recv gives me almost the same performance in term of Wtime. All > communications are postponed to MPI_Wait. > > I have tried calling mpi_test to push library to do communication during > iterations. And try to use a dedicated thread to do communication and the > other thread to do computing only. However, the performance gains are very > small or no gain at all. I'm wondering it is due to the hardware. The > cluster I tested uses 10G Ethernet card. > > > Best, > > Lei Shi > > > > > > On Fri, Apr 3, 2015 at 8:49 AM, Huiwei Lu <huiweilu(a)mcs.anl.gov> wrote: > >> Hi Lei, >> >> As far as I know, all current MPI implementations do not support true >> asynchronous communication for now. i.e., If there is no MPI calls in your >> iterations, MPICH will not be able to make progress on communication. >> >> One solution is to poll the MPI runtime regularly to make progress by >> inserting MPI_Test to your iteration (even though you do not want to check >> the data). >> >> Another solution is to enable MPI's asynchronous progress thread to make >> progress for you. >> >> -- >> Huiwei >> >> On Thu, Apr 2, 2015 at 11:44 PM, Lei Shi <lshi(a)ku.edu> wrote: >> >>> Hi Junchao, >>> >>> Thanks for your reply. For my case, I don't want to check the data have >>> been received or not. So I don't want to call MPI_Test or any function to >>> verify it. But my problem is like if I ignore calling the MPI_Wait, just >>> call Isend/Irev, my program freezes for several sec and then continues to >>> run. My guess is probably I messed up the MPI library internal buffer by >>> doing this. >>> >>> On Thu, Apr 2, 2015 at 7:25 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> >>> wrote: >>> >>>> Does MPI_Test fit your needs? >>>> >>>> --Junchao Zhang >>>> >>>> On Thu, Apr 2, 2015 at 7:16 PM, Lei Shi <lshi(a)ku.edu> wrote: >>>> >>>>> I want to use non-blocking send/rev MPI_Isend/MPI_Irev to do >>>>> communication. But in my case, I don't really care what kind of data I get >>>>> or it is ready to use or not. So I don't want to waste my time to do any >>>>> synchronization by calling MPI_Wait or etc API. >>>>> >>>>> But when I avoid calling MPI_Wait, my program is freezed several secs >>>>> after running some iterations (after multiple MPI_Isend/Irev callings), >>>>> then continues. It takes even more time than the case with MPI_Wait. So my >>>>> question is how to do a "true" non-blocking communication without waiting >>>>> for the data ready or not. Thanks. >>>>> >>>>> _______________________________________________ >>>>> discuss mailing list discuss(a)mpich.org >>>>> To manage subscription options or unsubscribe: >>>>> https://lists.mpich.org/mailman/listinfo/discuss >>>>> >>>> >>>> >>>> _______________________________________________ >>>> discuss mailing list discuss(a)mpich.org >>>> To manage subscription options or unsubscribe: >>>> https://lists.mpich.org/mailman/listinfo/discuss >>>> >>> >>> >>> _______________________________________________ >>> discuss mailing list discuss(a)mpich.org >>> To manage subscription options or unsubscribe: >>> https://lists.mpich.org/mailman/listinfo/discuss >>> >> >> >> _______________________________________________ >> discuss mailing list discuss(a)mpich.org >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/discuss >> > > _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] How to use non-blocking send/receive without calling MPI_Wait
by Lei Shi 03 Apr '15

03 Apr '15

Pavan, Thanks. You mean since I don't call MPI_Wait at all, the request object does not get a chance to be released by the MPI library? Right? I will give it a try. Thanks again! On Fri, Apr 3, 2015 at 10:12 AM, Balaji, Pavan <balaji(a)anl.gov> wrote: > > You can free the request with MPI_Request_free if you don't want to wait > on it. I have no idea how you'll write a correct program without waiting > for receive completions at least, though. > > -- Pavan > > > On Apr 2, 2015, at 11:44 PM, Lei Shi <lshi(a)ku.edu> wrote: > > > > Hi Junchao, > > > > Thanks for your reply. For my case, I don't want to check the data have > been received or not. So I don't want to call MPI_Test or any function to > verify it. But my problem is like if I ignore calling the MPI_Wait, just > call Isend/Irev, my program freezes for several sec and then continues to > run. My guess is probably I messed up the MPI library internal buffer by > doing this. > > > > > > On Thu, Apr 2, 2015 at 7:25 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> > wrote: > > Does MPI_Test fit your needs? > > > > --Junchao Zhang > > > > On Thu, Apr 2, 2015 at 7:16 PM, Lei Shi <lshi(a)ku.edu> wrote: > > I want to use non-blocking send/rev MPI_Isend/MPI_Irev to do > communication. But in my case, I don't really care what kind of data I get > or it is ready to use or not. So I don't want to waste my time to do any > synchronization by calling MPI_Wait or etc API. > > > > But when I avoid calling MPI_Wait, my program is freezed several secs > after running some iterations (after multiple MPI_Isend/Irev callings), > then continues. It takes even more time than the case with MPI_Wait. So my > question is how to do a "true" non-blocking communication without waiting > for the data ready or not. Thanks. > > > > > > _______________________________________________ > > discuss mailing list discuss(a)mpich.org > > To manage subscription options or unsubscribe: > > https://lists.mpich.org/mailman/listinfo/discuss > > > > > > _______________________________________________ > > discuss mailing list discuss(a)mpich.org > > To manage subscription options or unsubscribe: > > https://lists.mpich.org/mailman/listinfo/discuss > > > > _______________________________________________ > > discuss mailing list discuss(a)mpich.org > > To manage subscription options or unsubscribe: > > https://lists.mpich.org/mailman/listinfo/discuss > > -- > Pavan Balaji ✉️ > http://www.mcs.anl.gov/~balaji > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss > _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0