[OpenAFS] accessing R/O volume becomes slow

[Hans-Werner Paulsen]

On 11/26/2014 07:54 PM, Benjamin Kaduk wrote:
> On Wed, 26 Nov 2014, Hans-Werner Paulsen wrote:
>
>> Hello,
>> this is on Linux 3.14.8 x86_64, and OpenAFS 1.6.9. The machine is running
>> normally for several months, and then accessing a specific R/O volume (e.g. ls
>> -lR <large_volume>) becomes slow. Accessing the R/W version of this volume
>> works normally. Accessing other R/O volumes, which have the same size and
>> number of files, works normally. Accessing the R/O version of the problem
>> volume from other clients works normally.
>> The command "fs flushall" does not solve the problem. The (second) "ls -lR"
>> command needs 10 seconds on the R/O, and 2 seconds on the R/W version of this
>> volume.
>> Accessing the R/O version from other fileservers (using fs setserverprefs)
>> does not change anything.
>> Checking the machine I see more than 5 million of afs_inode_cache slab
>> entries. Is this normal? Any hint how to proceed?
> Are accesses from the same client to other volumes on the same (slow)
> fileserver slow or fast?  Maybe you are getting throttled for too many
> failed RPCs...
>
I have 2 volumes A and B (similiar in size and number of files) with R/W 
on machine X, and R/O on machine X, Y and Z. With "fs setserverprefs" I 
am accessing only the fileserver on X.
On the problem machine the first "ls -lR" takes about 1 to 2 minutes. 
The following "ls -lR"s take 4 to 5 seconds on A(R/W), B(R/W), B(R/O) 
and 15 seconds on A(R/O).
On another client machine all four values are identical and much better 
( 1.5 second).