Messages in this thread | | | Date | Tue, 29 Jun 1999 01:46:02 +0200 (CEST) | From | Andrea Arcangeli <> | Subject | Re: Perforamnce comparison between 2.3.8 and 2.2.10 |
| |
On Mon, 28 Jun 1999, Chuck Lever wrote:
>i also believe that a multithreaded page cache for file data won't help >much if the file system is still serialized :)
I think the filesystem and the page cache are different things and that it make tons of sense to have a threaded cache with a not scalable fs as having a threaded cache with a very slow HD. For example you may read from the page cache in parallel at 200mbyte/sec for CPU from a filesystem on an hd that goes a 1byte/sec :).
Usually when the fs is involved you always need to do some kind of read-I/O and so you are going to be slow and sleep anyway. Once you have your working set in memory, then you live enterely in the page/buffer cache and the fs code stops running completly (even the regular flush that you want to do to not lose too much data after a crash, runs locally in buffer.c without ask to the fs anymore since you just know where to write the dirty data on the blockdevice).
Andrea
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |