I asked:
> We've got two SS5 machines with 32megs of memory running Solaris 2.4
> with the tcx0 24bit color displays. The problem is that both machines
> are paging to disk furiously, all of the time. When we run "top" on
> these machines, we discover that the problem is that Xsun is taking up
> between 30 and 76 *megs* of memory. We're running the Openwindows
> window manager, since the X consortium doesn't have an X version that
> will support the tcx 24bit VRAM.
>
> Is anyone else having this problem with the 24bit color SparcStation 5s?
> If so, what was your solution? (Hopefully you won't tell me that I have
> to buy another 64megs of memory per machine... ) Thanks much.
Thanks to everyone who answered. The solution seems to be, both from
this list and from Sun, "live with it." The responses I got back:
* Its a 24 bit color, so Xsun will need 24 times the bits as a monochrome
display or 3 times an 8bit color display. All that extra memory usage
must be the backing store.
* Memory leaks in the Xsun process. Unfortunately I don't think it is,
since the size doesn't continue to grow. It just starts big and stays
that way.
* One person reported that she solved the problem by having Sun replace
the logic board.
* One person suggested trying "-bs -su" with the server invocation.
This didn't help.
* One person reported having this problem which went away when he applied
these 19 patches:
101242-08 101945-23 102007-01 102070-01 102922-01
101878-03 101959-03 102030-01 101879-01 101977-03
102044-01 102153-03 101880-03 102049-01 102216-01
101907-05 101981-02 102066-04 102336-01
I had most of these applied, and those that I didn't I've added. No
change.
* And finally, the comment from the Sun Engineer to whom I talked. He
said that with the modular design of the Solaris kernel, the kernel
is constantly loading modules from disk. This is the "constant paging"
sound that the users report. He maintains that this really is faster
than the /vmunix kernel of Solaris 1 fame. Secondly he says that the
size of Xsun reported by top and ps isn't really as big as it says- the
size reported is the potential size if all of the modules were loaded
at once. He then agreed that buying more memory for the machine may
speed it up, though. (This strikes me as the "It was broken when I
got here, honest! And my slingshot misfired anyway!" sort of argument,
but oh well)
The Sun engineer did report that there is an Xsun patch, 102057, but
he did not expect it to help. I am also completely unable to locate
that patch on any server. (They'd change me "time + materials" to
give me a copy). If anyone has a copy of this patch I'd be interested
to try it just to be complete...
So my conclusion: 24bit color on a Sparc 5 is an amazing memory hog,
and there isn't a thing I can do about it, except buy more memory.
Thanks to:
Ray Bagatella <rayb@nimrod.itg.telecom.com.au>
Paul Caskey <pcaskey@swcp.com>
Casper Dik <casper@fwi.uva.nl>
"Jeanne Gee" <jeanne@sudi.rdd.lmsc.lockheed.com>
Kevin.Sheehan@uniq.com.au (Kevin Sheehan {Consulting Poster Child})
Dan Stromberg - OAC-CSG <strombrg@hydra.acs.uci.edu>
+-----------------------------------------------------------------------+
| Christopher L. Barnard O When I was a boy I was told that |
| cbarnard@cs.uchicago.edu / \ anybody could become president. |
| (312) 702-8850 O---O Now I'm beginning to believe it. |
| http://cs-www.uchicago.edu/~cbarnard --Clarence Darrow |
| Cyber Rights Now: Accept No Compromise. |
+----------PGP public key available via finger or PGP keyserver---------+
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:10:21 CDT