[16364] in Athena Bugs
Re: sun4 8.2.10: emacs
daemon@ATHENA.MIT.EDU (John Hawkinson)
Tue Sep 29 10:07:30 1998
Date: Tue, 29 Sep 1998 10:07:27 -0400
To: Greg Hudson <ghudson@MIT.EDU>
Cc: bugs@MIT.EDU
In-Reply-To: "[16362] in Athena Bugs"
From: John Hawkinson <jhawk@MIT.EDU>
I suspect it would also be useful to obtain some information about
the file being editted. Perhaps whether the problem ocurrs with vi as
well as emacs, and if you change how emacs opens files (the whole
renaming versus save-in-place and soforth).
This rather strongly reminds me of an exchange from -i help on 15
September, which, to my regret, was never fully resolved. I attach
excerpts from the zephyr logs (help as well as help.d). I spent quite
a while hunting for an AFS problem, but I probably should have looked
more closely at emacs.
The executive summary is "this problem is not unique to this user",
but provides little new information.
--jhawk
(feeling somewhat guilty at this point)
Auth: yes Time: 10:39:09 Date: Tue Sep 15 1998 Host: w20-575-26
From: katherine c liu <richardl>
would anyone have any idea why it takes > 10 seconds for
me to save a file in emacs (c-x c-s) even though the file
is something like 50 lines long?
Auth: yes Time: 10:39:22 Date: Tue Sep 15 1998 Host: mushu
From: Hey, you, get off of my cloud
<pipa>
what sort of machine are you on?
Auth: yes Time: 10:39:31 Date: Tue Sep 15 1998 Host: mushu
From: Don't play with me, 'cause you'll play with fire
<pipa>
and what else are you running on it?
Auth: yes Time: 10:39:33 Date: Tue Sep 15 1998 Host: w20-575-26
From: katherine c liu <richardl>
sun ultra, in w20
Auth: yes Time: 10:39:44 Date: Tue Sep 15 1998 Host: w20-575-26
From: katherine c liu <richardl>
and i'm running communicator 4.05, and that's it
Auth: yes Time: 10:40:46 Date: Tue Sep 15 1998 Host: denmark-vesey
From: If your zwgc lived here, this packet would be home now <zeno>
I suspect just the typical AFS slowness.
Auth: yes Time: 10:40:50 Date: Tue Sep 15 1998 Host: mushu
From: Catch your dreams before they drift away...
<pipa>
hrm, doesn't sound like the machine is particularly
hosed... that would be my explination. Either that or the
network connections are being slower in the cluster...
Auth: yes Time: 10:41:11 Date: Tue Sep 15 1998 Host: x15-cruise-basselope
From: John Hawkinson, KB1CGZ <jhawk>
What on earth is "typical AFS slowness"?
Auth: yes Time: 10:41:41 Date: Tue Sep 15 1998 Host: denmark-vesey
From: If your zwgc lived here, this packet would be home now <zeno>
You know, John... it's sluggish. Always has been.
Auth: yes Time: 10:42:03 Date: Tue Sep 15 1998 Host: x15-cruise-basselope
From: John Hawkinson, KB1CGZ <jhawk>
No, I don't, and I don't know what "sluggish" means? Can you give
a rigorous definition that isn't subjective?
Auth: yes Time: 10:44:33 Date: Tue Sep 15 1998 Host: denmark-vesey
From: If your zwgc lived here, this packet would be home now <zeno>
I can't, rigorously. This is really just an impressionistic
sort of thing. AFS has always seemed rather slow. Ever
done folder -pack on a 2000-message folder? It
takes *forever*, compared to relatively snappy to do
the same thing on a local (UFS) directory. I imagine
this is mostly just the network back-and-forth transit time.
Auth: yes Time: 10:46:00 Date: Tue Sep 15 1998 Host: x15-cruise-basselope
From: John Hawkinson, KB1CGZ <jhawk>
Right, but that's an operation that requires dealing with a lot of
individual small files, something that AFS is particularly weak at,
because it's caching paradigm can't do much there.
That's very much a special case and not a normal operation.
It feels particulary irksome to me to have people dismiss things as
"Oh, that's just <foo> being slow" because as often as not either
<foo> *shouldn't* be slow and people are ignoring a real problem,
or the assumption that <foo> is slow is just plain wrong, too. Without
a reasonable benchmark or test, it's hard to say such a thing.
Auth: yes Time: 10:48:09 Date: Tue Sep 15 1998 Host: denmark-vesey
From: If your zwgc lived here, this packet would be home now <zeno>
I dunno, I long ago accepted it as a fact of AFS life, so
to speak. Even just saving one AFS file, as in the original
question, takes a perceptibly longer time than does a local
file. Granted, >10 sec does sound extreme; I was probably
being too facile in attributing his (her? sorry, don't
remember) particular case to AFS.
Auth: yes Time: 10:49:06 Date: Tue Sep 15 1998 Host: x15-cruise-basselope
From: John Hawkinson, KB1CGZ <jhawk>
So, moving away from the meta-discussion for a bit,
Katherine, where is the file you were saving that took >10seconds
located? In your locker?
Auth: yes Time: 10:49:18 Date: Tue Sep 15 1998 Host: w20-575-26
From: katherine c liu <richardl>
yup
Auth: yes Time: 10:49:38 Date: Tue Sep 15 1998 Host: x15-cruise-basselope
From: John Hawkinson, KB1CGZ <jhawk>
Is it reproducible? Was it a one-time event or does it continue?
Auth: yes Time: 10:50:12 Date: Tue Sep 15 1998 Host: w20-575-26
From: katherine c liu <richardl>
it's reproducible.
i'm writing a program in c, and every time i make a change
and need to save it, it takes awhile.
gcc runs really fast, though, and top tells me i'm 99% idle and
xload doesn't have any significant peaks.
Auth: yes Time: 10:54:08 Date: Tue Sep 15 1998 Host: buzzword-bingo
From: Russell S Schwartz <rsschwar>
richardl: Is Communicator doing anything? It can use a lot of memory, so
maybe it is causing parts of emacs to swap out that need to be swapped back
in when you save. This is just a not very informed guess, though.
Auth: yes Time: 10:54:40 Date: Tue Sep 15 1998 Host: tux
From: You are in a maze of twisty little zsigs, all alike. <amu>
Some versions of Netscape also like to eat CPU for no good
reason.
Auth: yes Time: 10:55:12 Date: Tue Sep 15 1998 Host: w20-575-26
From: katherine c liu <richardl>
exiting communicator doesn't seem to help, and it wasn't doing
anything to begin with. just sitting on a web page (with no java,
no animations, no dhtml, ntohing like that).
Auth: yes Time: 11:10:34 Date: Tue Sep 15 1998 Host: x15-cruise-basselope
From: John Hawkinson, KB1CGZ <jhawk>
I spent some time on-and-off trying to look at this and find no useful
data. grr.
Auth: yes Time: 11:11:16 Date: Tue Sep 15 1998 Host: tux
From: We're Certainly Dwarves. <amu>
this being alleged AFS performance issues?
