Inefficient swapping with swapfile?

Inefficient swapping with swapfile?

Post by Alexander Hin » Sun, 06 Aug 1995 04:00:00



I have set up a 16 meg swapfile on my 16meg linux system running X and boy does this baby have abysmal performance.  I can barely run netscape and emacs at the same time, before I get into the hellishly long swapfest.  It seems that when I execute free, that the swap space is not really filling up and the os seems to be playing with its buffers.  I'm not really sure what the heck is going on.(I'm running UMSDOS, but still...)

I haven't seen this in a FAQ anywhere.  Can anybody please give me some pointers about imroving this situation?  Thanks in advance.

--X

 
 
 

Inefficient swapping with swapfile?

Post by Billy Ch » Tue, 08 Aug 1995 04:00:00


Alexander> I have set up a 16 meg swapfile on my 16meg linux system
Alexander> running X and boy does this baby have abysmal performance.
Alexander> I can barely run netscape and emacs at the same time,
Alexander> before I get into the hellishly long swapfest.  It seems

A swap partition is much faster than a swapfile.  Something to do with
the need to handle the file system?

/Billy

 
 
 

1. Inefficient swapping algorithm?

Hi there,

I guess that this is really something to be read by mr. Torvalds,
and I guess he's
not a member of this group, but I haven't got his address with me---the
mosaic Linux home page doesn't answer either.

I have been observing the usage of the swapspace on my linux config
(I've got a 486/33*2, 16 Meg swap, 4 Meg mem->a low-budget configuration)
and did the following exeriment:

in one virtual console: primes 1 | less
in an other VC: top

I paged through the primes until it was decided that less should be swapped
out partially. (This is at approx byte 10^6, top showed 2*10^6 mem used)
Then, on going to the first line with less, a region of memory that has just
been swapped out. Hard disk lits, makes hardly any noise -> it's swapping all
right. The reaction time is quite slow, it is swapping for about 1 sec,
so I guess about 1 Mb of memory is swapped (my HD transfer rate is
above 1Mb/sec). This, however is not true. If I page through the primes again,
I reach byte ~1000 soon, and the machine starts swapping again. (I have,
again, reached the `oldest' part of memory again---that is swapped out first.)
Again, it takes more than a second to finish the swapping. At the next 1k
boundary of memory, there will be swapped again, etc. Apparently, a swap
page is 1k in size, which makes sense. But I don't understand why swapping
1k in and 1k out takes that long!

It appears to me that the swap space is allways kept `in order,' some how,
or that something else is happening. From the noise, the space is really
read/written in consecutive blocks,
as a swapspace might better be (I don't know),
but 1 Mb (or more) disk i/o for a single kb of swapped memory seems very
inefficient. From the experiments I did not get any clue that actually more
than 1kb-at-the-time was actually been swapped.

Question: does anybody know about swap strategies?


P.S. As a result, X is *very* slow. I know I should buy more memory,
but as long as prices are determined by speculation and other maffia-like
reasons, I'd like to stall that for a while.

2. shared memory amoung child processes

3. hardrive removed, including swapfile how to stop mounting swapfile and make another

4. can linux do something that windows can't?

5. Swapfile vs. Swap partition

6. : file system error?

7. Why is a swap partition better than a swapfile?

8. bug, error, pb de linkage ?

9. Using swapfile instead of swap partition

10. Is LiveConnect inherently inefficient?

11. Modules too inefficient?

12. Why so inefficient source RPM's ??

13. Solaris uiomove() inefficient?