[OpenAFS] 1.3.6000 integrated logon?

Kevin Coffman kwc@citi.umich.edu
Fri, 19 Mar 2004 10:43:58 -0500


This is a multi-part message in MIME format.

------=_NextPart_000_000E_01C40D9F.2C0A9A30
Content-Type: text/plain;
	charset="US-ASCII"
Content-Transfer-Encoding: quoted-printable

I'm at home today, so I can't do much to help diagnosing.  The problem I =
saw
was a 30-60 delay after double-clicking the tray icon before the dialog
appeared.  Switching between tabs from tokens to drive mappings takes =
30-60
seconds or more.  Redrawing the dialog box has a similar delay.  I =
didn't do
a network trace, but I did look at CPU.  afscreds is using CPU, but it's =
not
like it is pegged (2 to 9% usage).

-----Original Message-----
From: Jeffrey Altman [mailto:jaltman@columbia.edu]=20
Sent: Friday, March 19, 2004 9:42 AM
To: Stephen Joyce
Cc: Kevin Coffman; openafs-info@openafs.org
Subject: Re: [OpenAFS] 1.3.6000 integrated logon?


Stephen Joyce wrote:


FWIW, I've seen that very slow behavior too, but figured it was =
something

I'm doing wrong... (the new versions, .5299 and .6000, also seem to

consistently crash, with the question about sending a crash report to =
MS,

every time I stop the service.)  I see this on WinXPsp1 with only MS

patches and OpenAFS installed.






First, the crash is most likely in the new power
management thread.  The thread is not always=20
shutting down before the service control thread
finishes.  I thought I had that fixed.


As for the slow dialogs I am not sure.  Is there
any additional information you can provide? =20

What do you mean by slow?  Delays while entering
text?  Delays from the time the "OK" button is
pressed until the tokens are obtained? =20

One point of information is that in OpenAFS I=20
discovered a very bad race condition between=20
processes or threads on the same machine within
ktc_SetToken() or ktc_GetToken() calls.  Each
call must perform two steps a pioctl() and an
RPC.  It is possible for two processes making
calls at the same time to be stair stepped=20
resulting in undesireable results.  Therefore,
I added a global mutex around the pairs of calls
to ensure that only one thread/process is=20
attempting to perform an operating at a time.
This will introduce a very slight delay because
the Obtain Tokens thread is competing with=20
another thread within afscreds.exe which is
also performing the pioctl()/RPC call combinations
to obtain token status information.  However,=20
I cannot imagine the delay is all that great.

Does a network trace reveal anything interesting?

Jeffrey Altman





------=_NextPart_000_000E_01C40D9F.2C0A9A30
Content-Type: text/html;
	charset="US-ASCII"
Content-Transfer-Encoding: quoted-printable

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD>
<META HTTP-EQUIV=3D"Content-Type" CONTENT=3D"text/html; =
charset=3Dus-ascii">
<TITLE>Message</TITLE>

<META content=3D"MSHTML 6.00.2800.1400" name=3DGENERATOR></HEAD>
<BODY text=3D#000000 bgColor=3D#ffffff>
<DIV><SPAN class=3D144373715-19032004><FONT face=3DArial color=3D#0000ff =
size=3D2>I'm at=20
home today, so I can't do much to help diagnosing.&nbsp; The problem I =
saw was a=20
30-60 delay after double-clicking the tray icon before the dialog=20
appeared.&nbsp; Switching between tabs from tokens to drive mappings =
takes 30-60=20
seconds or more.&nbsp; Redrawing the dialog box has a similar =
delay.&nbsp; I=20
didn't&nbsp;do a&nbsp;network trace, but I did look at CPU.&nbsp; =
afscreds is=20
using CPU, but it's not like it is pegged (2 to 9% =
usage).</FONT></SPAN></DIV>
<BLOCKQUOTE=20
style=3D"PADDING-LEFT: 5px; MARGIN-LEFT: 5px; BORDER-LEFT: #0000ff 2px =
solid; MARGIN-RIGHT: 0px">
  <DIV></DIV>
  <DIV class=3DOutlookMessageHeader lang=3Den-us dir=3Dltr =
align=3Dleft><FONT=20
  face=3DTahoma size=3D2>-----Original Message-----<BR><B>From:</B> =
Jeffrey Altman=20
  [mailto:jaltman@columbia.edu] <BR><B>Sent:</B> Friday, March 19, 2004 =
9:42=20
  AM<BR><B>To:</B> Stephen Joyce<BR><B>Cc:</B> Kevin Coffman;=20
  openafs-info@openafs.org<BR><B>Subject:</B> Re: [OpenAFS] 1.3.6000 =
integrated=20
  logon?<BR><BR></FONT></DIV><FONT face=3D"Bitstream Cyberbit">Stephen =
Joyce=20
  wrote:<BR></FONT>
  <BLOCKQUOTE =
cite=3DmidPine.GSO.4.58.0403190841340.9669@titan.physics.unc.edu=20
  type=3D"cite"><PRE wrap=3D""><FONT face=3D"Bitstream Cyberbit">FWIW, =
I've seen that very slow behavior too, but figured it was something
I'm doing wrong... (the new versions, .5299 and .6000, also seem to
consistently crash, with the question about sending a crash report to =
MS,
every time I stop the service.)  I see this on WinXPsp1 with only MS
patches and OpenAFS installed.

</FONT></PRE></BLOCKQUOTE><BR>First, the crash is most likely in the new =

  power<BR>management thread.&nbsp; The thread is not always =
<BR>shutting down=20
  before the service control thread<BR>finishes.&nbsp; I thought I had =
that=20
  fixed.<BR><BR><BR>As for the slow dialogs I am not sure.&nbsp; Is =
there<BR>any=20
  additional information you can provide?&nbsp; <BR><BR>What do you mean =
by=20
  slow?&nbsp; Delays while entering<BR>text?&nbsp; Delays from the time =
the "OK"=20
  button is<BR>pressed until the tokens are obtained?&nbsp; <BR><BR>One =
point of=20
  information is that in OpenAFS I <BR>discovered a very bad race =
condition=20
  between <BR>processes or threads on the same machine =
within<BR>ktc_SetToken()=20
  or ktc_GetToken() calls.&nbsp; Each<BR>call must perform two steps a =
pioctl()=20
  and an<BR>RPC.&nbsp; It is possible for two processes making<BR>calls =
at the=20
  same time to be stair stepped <BR>resulting in undesireable =
results.&nbsp;=20
  Therefore,<BR>I added a global mutex around the pairs of calls<BR>to =
ensure=20
  that only one thread/process is <BR>attempting to perform an operating =
at a=20
  time.<BR>This will introduce a very slight delay because<BR>the Obtain =
Tokens=20
  thread is competing with <BR>another thread within afscreds.exe which=20
  is<BR>also performing the pioctl()/RPC call combinations<BR>to obtain =
token=20
  status information.&nbsp; However, <BR>I cannot imagine the delay is =
all that=20
  great.<BR><BR>Does a network trace reveal anything =
interesting?<BR><BR>Jeffrey=20
  Altman<BR><BR><BR></BLOCKQUOTE></BODY></HTML>

------=_NextPart_000_000E_01C40D9F.2C0A9A30--