Messages in this thread Patch in this message | | | Date | Wed, 25 Nov 1998 15:27:49 GMT | From | "Stephen C. Tweedie" <> | Subject | Re: MM tuning, was: Re: Linux-2.1.129.. |
| |
Hi,
On 21 Nov 1998 11:15:35 +0100, Andi Kleen <ak@muc.de> said:
> I'm not Stephen, but... (AFAIK Stephen is away in Holland)
Boo!
> This last patch of Stephen is still required stop the 'free half of my > memory when a lot of processes are active and freepages.low is reached' > (this happens when compiling the kernel with make -j4).
Linus, I know this particular check feels perhaps a bit more arbitrary than the others, but I really think it is necessary, primarily because we have got a solid, reliable test case which proves that under a not-uncommon load, we repeatedly swap out insanely large amounts of memory. Andi has been seeing tens of MB (up to 40 or 50MB) of overswap without this patch in place when doing parallel makes, simply because of the large numbers of processes which start swapping out at once.
Other than that, the 130-pre3 vm is looking good (now if only I could work out why seyon on /dev/ttyS2 reliably locks this box up solid).
--Stephen
Index: linux/mm/vmscan.c =================================================================== RCS file: /vger/u4/cvs/linux/mm/vmscan.c,v retrieving revision 1.56 diff -u -u -r1.56 vmscan.c --- vmscan.c 1998/11/18 07:17:02 1.56 +++ vmscan.c 1998/11/21 09:12:37 @@ -570,6 +583,10 @@ if (!retval) break; count--; + if (!(gfp_mask & __GFP_DMA) && + (nr_free_pages > freepages.high + SWAP_CLUSTER_MAX + || atomic_read(&nr_async_pages) > SWAP_CLUSTER_MAX)) + break; } while (count > 0); current-> flags &= ~PF_MEMALLOC; }
- 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/
| |