Messages in this thread | | | Date | Thu, 26 Nov 1998 10:03:30 -0800 | From | Benjamin Redelings I <> | Subject | Re: [2.1.130-3] Page cache DEFINATELY too persistant... feature? |
| |
Thanks for all the responses! So its writes that are causing the problem.
> In some situations this is a good readout. If the O/S is doing its job, > then that 24Mb in swap is disused code and data pages that are rarely > accessed, if at all. As a bonus, that 24Mb can now be used for cache and > buffers to reduce disk accesses and really make the system fly. Good point. In fact, the paged-out pages were mostly unused pages, I think, because xosview did not show a high rate of page-ins. However, due to the sheer AMOUNT of pages that were paged out, a large bit of netscape got paged out also, which I was actually using. So it had to page a bit before the scrollbar worked, then some for the back button, etc :). The problem is not so much that the swapping was bad, as the AMOUNT of swapping that was done; the page-cache should have been shrunk also, I think. Usually, though, after recovering from such an episode, the system feels quite nice with 9.5Mb or so swapped to disk.... evidently thats the portion of memory that wasn't used much.
> I agree a "free" report like the above can be alarming. However, the > important things you need to compare with previous releases are > a) How snappy interactive performance felt > b) How quickly the "background" task of updating your Debian system > completed b) So, the upgrade that was causing the swapping took a bit longer.The fact is the page-cache simply wasn't being used. It was just sitting there taking up 33% of my RAM. So, normally, disk usage is reserved for writes instead of writes plus paging in and out. a) Interactive performance went DOWN. Netscape had to do some page-ins to respond to anything :)
-BenRI
P.S. I take it that its OK to Cc people like this even if they're on linux-kernel? Or is that bad form?
-- If E. Coli was a computer: "OPERON.DLL: Mutation in LACTOSE.DLL. Entire system halted."
Benjamin Redelings I <>< http://sdcc13.ucsd.edu/~bredelin
- 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/
| |