Based on the information provided there isn't a good way to answer why the creation time is different in the one to two client case and in comparison to the results in a paper. I would guess there are two possibilities:
1) The server and network hardware is different between your tests and the tests that achieved 700 creates/second with two servers. 2) The modifications you have made to pvfs2-touch.c adds overhead to the file creations process It's not easy to track down where the added latency is coming from. I would recommend enabling various levels of logging or using traffic captures to get an idea where the extra latency is occurring. Good luck, Michael 2011/9/21 欧佳欣 <[email protected]> > In orangefs-2.8.3-EXP, I modified the interface of pvfs2-touch.c , > after modified , when I use the command pvfs2-touch A , it will create 10000 > files(A1,A2,.....A10000). > > Then I installed the orangefs on a cluster that has 3 MetaData > Servers , 1 Data Server , and 2 Clients. and no two of these servers are on > the same machine . And I conducted two different types of tests , just as > follows : > > Firstly I only use one client to create 10000 files , and it costs > about 280s. > > Sencondly, I use mpi to let the two clients create 10000 files > each at the same time , and it costs about 330s. > > I have seen a paper before , that talk about the average file > creation throughput with different servers in pvfs2 , and it said that the > throughput of pvfs2 is about 700 creates per seconds on two servers and > about 1100 creates per seconds on three servers.But in my test, it only has > the throughtput of no more than 100. > > So I want to know why ? > > _______________________________________________ > Pvfs2-developers mailing list > [email protected] > http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers > >
_______________________________________________ Pvfs2-developers mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
