Re: fio: current verify status

On 7/30/07, Jens Axboe <jens.axboe_at_oracle.com> wrote:
>
> Hi Shawn/list
>
> I cleaned up and committed some verify changes today, basically:

Thanks for the clean ups! I took a look at the changes and they all look
good.

- Renamed header_offset and header_interval options to verify_offset and
> verify_interval instead. I think it better represents what the options
> do.

Ya I was trying to separate the concepts of writing a block header and
actually doing verification, but it's not really necessary.

- Added verify_pattern option. I hope this does what you want, if not
> then feel free to modify it!

This looks awesome :)

- Split the checksum data from the verify header. For some of the
> smaller checksums, we "waste" too much of the data block size because
> the verify_header encompasses everything from crc7 to sha512.
>
> - Moved the crc stuff to crc/
>
> I'm done with the cleanups I wanted to do, so I wont step on your code
> anymore for the time being! So from your list, what is still pending is:
>
> - Verify without IO log. This will be very nice, looking forward to
> that. I'm assuming this also means that you can run a read verify of a
> file without knowing the block sizes from when it was written? That
> would be useful. I'm assuming that is has to handle that, otherwise I
> don't see how verify without an IO log and bsrange= would work.

Can't we reproduce the random sequence by saving the seeds used to generate
it?

Also verify_offset probably doesn't currently play nice with bsrange=. I'm
less concerned with bsrange= but I'll try to make io-log-less verification
work for as many cases as I can.

- The extra signature containing info about the block written. Also a
> very useful addition, I think.
>
> So I hope you'll proceeded and get those in. Thanks!
>
> --
> Jens Axboe
>
>
Received on Tue Jul 31 2007 - 04:20:50 CEST