Messages in this thread | | | Subject | Re: 2.6.1-rc1-tiny2 | From | Trond Myklebust <> | Date | Wed, 07 Jan 2004 16:41:38 -0500 |
| |
På on , 07/01/2004 klokka 15:10, skreiv Matt Mackall: > NFS is a good example of why the guarantees of mempool are being > overstated - it still needs to allocate SKBs to make progress and > preallocating a pool for other data structures can make that fail > where it otherwise might not. The pool size for NFS (32) is also > completely arbitrary as far as I can tell.
If you are in a hardware situation where you actually care about the permanent size of that mempool, then you're barking up entirely the wrong tree: there is a hell of a lot more memory to reclaim from not having to build up all those nfs_page lists in the first place.
i.e. Rip out the entire asynchronous NFS read/write support, not just the mempools.
As for the usefulness of the mempools in the situation where you have asynchronous I/O: I agree that the socket layer screws any chance of a guarantee. So does the server if it goes down, the network itself can screw you,.... All in all, it is surprising how few guarantees NFS offers you. I therefore see the mempools as more of an optimization that mainly avoid sleeping under a certain limited set of "reasonable" circumstances.
Cheers, Trond - 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/
| |