Conclusions

migShm enables the migration of certain shared memory applications, such as Apache 1.3.X and Blast. We also know that the migShm patch does not enable pthread migration yet.

However There are some known issues with the current migShm patch. First of all it does not work fine in XFree environments and some window managers. One could argue that clusternodes do not need XFree, but one then forgets about the cohabitation of openMosix and LTSP in some environments.

The second problem is that for some modules , e.g. some SCSI modules, the kernel does not build anymore after applying the openMosix+migShm patch, the kernel does not build anymore after applying the openMosix+migShm patch. MAASK group would appreciate some help in fixing this problem.

Future developments of migShm might inlcude implementations such as the splitting of shared memory, where a certain process accesses certain regions of the shared memory more frequently only partial memory migration might be required. Also the migration of a group of processes that use the same memory might be possible.

As we have seen through these tests, stability is a major issue , I would not consider the migShm patch as ready for production yet, as where the openMosix patch has already proven that is ready for prime time clustering. We need people to test these developments over and over again and also send in bug reports to the Maask group. It is only with these bugreports that we can improve the stability of the patch. Testing should also be done on a variety of network environments. Given the fact that migrating shared memory processes uses more bandwith than migrating other procesess, a 10Mb network connection was probably far from ideal. Hence a conclusion could be not to run migShm on these kind of networks.

Next to the fixing of these bugs there still is some other work to do. But as soon as this patch is stable enough it might have a chance of becoming a part of the default openMosix patch.