Messages in this thread | | | Date | Tue, 31 Mar 1998 18:01:39 +0530 (GMT+0530) | From | Chirayu Patel <> | Subject | Re: Magic in swap code. :-) |
| |
On Mon, 30 Mar 1998, Rik van Riel wrote:
> On Fri, 27 Mar 1998, Chirayu Patel wrote: > > > Please let me have your comments on it. I am working on a patch for > > 2.1.90, which is almost complete. > > It would be best to reserve the SWAP_READONLY behaviour > for swapping-from-a-tasks-context, since we don't really > care about kswapd blocking for disk I/O... > > I'll test it and implement the swapping-RO-from-user-context > idea at the same time... >
I have a 16MB machine. What happens is that whenever I have used up a considerable amount of swap space...and say I open a Netscape preference window. It takes a lot of time and there is a lot of hard-disk activity and my mouse control becomes jerky.
What I feel is happening is - kswapd does a asychronous write of the dirty pages to the disk and frees the required number of pages. But by the time the Netscape preference window gets loaded in memory we have already started writing the dirty pages to the swapdevice..and this increases the load in the hard disk.
My fix proposes to delay the freeing of the dirty pages to the disk ..and only do the writes (asynchronously ofcourse!) when the system is IDLE or enough read only pages are not available.
I have put the patch for this in my system and I am observing a noticable improvement in the interactive-ness.
The other thing which I had proposed is to determine if the system is sufficiently IDLE we can pre-swap the dirty pages and reset the dirty bit. So even these pages can be swapped out (as they are no longer dirty) when kswapd executes. Ofcourse the sort of behaviour depends on a lot of parameters as described in my previous mail.
Comments please.......
-- Chirayu
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu
| |