[Storage-research-list] File-system traces

[Jeff Chase]

Forgive me if this is off-topic, but it seems like a good time and place 
to put in a plug for Darrell Anderson's fstress package, which is a 
synthetic trace generator and load generator for NFSv3/UDP.   It 
generates NFS traces from a set of input parameters that define 
reference distributions etc.  etc.  If your goal is to learn something 
from the traces, then it won't help you.  But if your goal is to 
evaluate storage servers then it can be very helpful. 

http://www.cs.duke.edu/ari/fstress/

Fstress has canned settings for several "representative" NFS workloads, 
including SPECsfs.  If and when a larger set of NFS traces does become 
available, it would be useful to extract the parameters to generate a 
"similar" workload synthetically, e.g., with fstress.  Of course this is 
a research topic.

Fstress has been quietly aging, but we have recently started using it 
again and it seems to be pretty robust.   A good group of folks have 
been using in industry, but I still see some papers go by from the 
research community with really hoary evaluation methodologies like "the 
Andrew benchmarks" or much more limited workload gens like Postmark.  As 
a community I think we'd be better off with evaluations based on fstress 
until something better comes along.

Jeff

Greg Ganger wrote:
>
> Of course, many of us share your hope.  Please note that IBM is 
> absolutely
> welcome to contribute NFS traces (or any other I/O traces) to this 
> community
> repository.
>
> Greg
>
>
> On Tue, 8 May 2007, Ohad Rodeh wrote:
>
>>
>> List,
>>   Thanks for all the responses I got. I had made a mistake in asking for
>> the cello traces, I actually need a file-system, not block-level, 
>> trace. I
>> stand corrected. Anyhow, I have one comment, the SNIA repository at
>> http://iotta.snia.org/traces/ does not contain any NFS traces 
>> currently. I
>> hope that will change in the future.
>>
>>      Ohad.
>>
>> _______________________________________________
>> Storage-research-list mailing list
>> Storage-research-list at ece.cmu.edu
>> https://sos.ece.cmu.edu/mailman/listinfo/storage-research-list
>>
> _______________________________________________
> Storage-research-list mailing list
> Storage-research-list at ece.cmu.edu
> https://sos.ece.cmu.edu/mailman/listinfo/storage-research-list