![]() | |||||||||||||
Messages in this thread |
Hubertus Franke <frankeh@watson.ibm.com> writes: > > Agreed.. here are some issued we learned from other projects that had > similar interception points. > > Having a central umbrella object (let's stick to the name container) > is useful, but being the only object through which every access has to > pass may have drawbacks.. > > task->container->pspace->pidmap[offset].page implies potential > cachemisses etc. > > If overhead becomes too large, then we can stick (cache) the pointer > additionally in the task struct. But ofcourse that should be carefully > examined on a per subsystem base... Ok. After talking with the vserver guys on IRC. I think grasp the importance. The key feature is to have a place to put limits and the like for your entire container. Look at all of the non-signal stuff in struct signal for an example. The nested namespaces seem to be just an implementation detail. For OpenVZ having the other namespaces nested may have some importance. I haven't gotten their yet. The task->container->pspace->.... thing feels very awkward to me, and feels like it increases our chance getting a cache miss. So I support the concept of a place to put all of the odd little things like rlimits for containers. But I would like to flatten it in the task_struct if we can. Eric - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ | ||||||||||||
| Last update: 2006-02-08 05:08 [W:18.413 / U:0.090 seconds] ©2003-2008 Jasper Spaans | |||||||||||||