Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On 06/27/2011 10:20 AM, Felix Frank wrote:
> On 06/26/2011 05:30 PM, Andreas Hofmeister wrote:
>> With 'dd' on a system with similar performance (8x15k SAS disks, LSI,
>> 10GBE), I get 1GByte/s into the local block device but only 600MByte/s
>> into the DRBD device. Using another benchmark ("fio" with async-I/O and
>> larer values for I/O-depth), I can get up to about 1.2GByte/s in both
>> cases.
>
> Yes, dd is not a typical workload. It can be worthwile to use benchmarks
> with more "realistic" workloads, but frankly, I disbelieve that many
> people will find tools that accurately model their respective target
> workloads in that respect.
>
> I like the fact that dd will report a specific technical boundary to me.
> If my real life workload load leads to even higher throughput, that is
> appreciated.
>
> An interesting question for me is: Is tuning for dd-like workloads
> potentially harmful to random access throughput?
> I certainly don't hope so.
What I would like to see is a set of standardized configurations for "fio"
(or similar tools) that cover some basic common workloads. That would allow
for more realistic benchmarking yet allow a wide comparison of results. I
imagine a site with a list of named configurations with an explanation of
why these parameters were chosen and what kind of use-case it is supposed
to emulate. For bonus points one could add the ability to post system
configurations with results and allow comments.
Regards,
Dennis