[OpenAFS] Windows cache rehashed...

[Rodney M Dyer]

Jeffrey,

I downloaded and tested OpenAFS 1.3.51 using the same test I ran with 
1.2.10.  The results are not encouraging.  In fact, the results show that 
it fails in less time and with fewer files.

http://www.coe.uncc.edu/~rmdyer/test_8MB_afscache1351.jpg

This screenshot shows the results after I let the machine sit over the 
weekend.  The handles did not drop as expected.

I must say I was rather dazed (blindsided) by the new 1.3.51 client.  There 
are a good many operational characteristics changed from previous 
versions.  I noted the following...

1.  A "FreelanceClient" setting.  Can someone explain what this is?  When 
it is enabled, the client doesn't function properly.  Navigating down the 
AFS tree of our cell results in...

      "n:\uncc\uncc\uncc\uncc\uncc..."

      I had to disable the setting to get the client to function properly.

2.  An associated "FreelanceClient" file called 
"c:\winnt\afs_freelance.ini".  What is the purpose of this file?

3.  The historical operational service name is changed from "IBM AFS 
Client" to "OpenAFS Client Service".

4.  The historical software directory tree has been relocated to 
"c:\Program Files\OpenAFS" instead of "c:\program files\IBM\AFS".

5.  The installer has changed (expected from what I've been reading in this 
list).

Anything else I should know about?  This could also be a shock to anyone 
who has used the AFS client for as long as we have, or have scripts and 
software tools that expect things to be in the old locations.  There's very 
little hint on the OpenAFS web site about these legacy changes.

I don't disagree with the changes, I was just a bit shaken by them.

Rodney

At 03:50 PM 12/19/2003, Jeffrey Altman wrote:
>Since development on the 1.2 branch has stopped and because the last two
>
>months have been spent by Rob Murawski, Joe Beuhler, and myself finding
>and fixing memory and handle allocation issues, it would be nice if you
>could perform your testing using the current development builds which I
>have pointed this list to over the last month.  You can now find a link
>to the latest development build at
>
>   http://www.openafs.org/release/latest-unstable.html
>
>It would really suck if the problems had been fixed in the development
>tree before this thread began.
>
>As for the large number of handles being allocated, that is not
>necessarily a bug unless the handle count does not descrease over time
>as the cache information is no longer needed.  In my earlier testing the
>
>handle count went over 800 but dropped all the way back down over time
>as AFS use dropped off.  In my earlier testing, I used a single 600MB
>file.  I will perform a new test with directory trees.
>
>Jeffrey Altman
>
>
>Rodney M Dyer wrote:
>
> > To reproduce the problem, use the following settings...
> >
> >      Windows XP SP1, 1 Gig RAM, P4 3.0 Gig, 100 MBit connectivity
> >      OpenAFS 1.2.10
> >      Cache size:  8192K  ( 8 Meg cache )
> >