Messages in this thread | | | Date | Tue, 4 May 1999 21:37:46 +0200 (CEST) | From | Andrea Arcangeli <> | Subject | Re: [patch] swap cache read accesses |
| |
On Tue, 4 May 1999, Stephen C. Tweedie wrote:
>Hi, > >On Tue, 4 May 1999 13:31:33 +0200 (CEST), Andrea Arcangeli ><andrea@e-mind.com> said: > >> I was running `./bw_mem 120m rd' (from lmbench) and I noticed my swap to >> be eaten completly in some second. All the cache was full of swap cache >> pages. > >> The problem is that there is no good reason to map in the process a >> read-only swap cache page if the swap cache page was not shared. > >Yes there is. That's the whole reason the swap cache was added in the >first place. It means that if we want to reuse the process's physical >memory again, we don't have to write it back to swap.
Ah yes. Thanks. Now I am using HINT hacked by me to work only in the swap and it gives better performances dropping the swap cache directly (ask me privately to see the graph).
>It will cause a page fault, yes, but not a COW: the do_wp_page will just
Yes, it doesn't do a COW, but it has a cost anyway (and right now we get_free_page and free_page in the no-COW case too).
I'll reput the !write_acess back here too just because seems to make tons of sense, even if with such logic my 70mbyte of swap are too low for 128mbyte of RAM, at least while running the bw_mem rd test... :(
Thanks.
Andrea Arcangeli
- 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/
| |