[OpenAFS] afs.GCPAGs in current releases under Linux (RHEL4/5)

Simon Wilkinson sxw@inf.ed.ac.uk
Fri, 5 Mar 2010 18:48:53 +0000


On 5 Mar 2010, at 18:34, Eric.Hagberg@morganstanley.com wrote:

> This took a little longer to set up than I'd hoped, as 1.5.72  
> doesn't work under RHEL4 (the platform on which I was doing most of  
> my tests), due to the lack of zero_user_segments and page_offset in  
> the compiled kernel module.

Okay, I should fix both of these (I think they're both my fault).

> However, the results running 1.5.72 on a RHEL5 system are pretty  
> much the same as with 1.4.x (where x is greater than or equal to 2),  
> wrt system time creep.
>
> One thing I'd noticed in RHEL4 tests (systemtap doesn't seem to give  
> the same result under RHEL5) is that the system time usage  
> corresponds to high frequencies of calling afs_ComputePAGStats.  
> Compiling with AFS_NOSTATS defined or commenting out the section of  
> code in src/afs/afs_user.c that calls afs_ComputePAGStats also  
> relieves the problem.

Just to be clear here, are you seeing a high number of calls to  
afs_ComputePAGStats, or a large amount of system time spent in that  
function?

I can't see a call to afs_ComputePAGStats in afs_user.c - are you just  
commenting out the body of the function, or is there a call site that  
I am missing?

Cheers,

Simon.