[Iometer-devel] RE: Iometer changes (individually packaged) to run 'dynamo' onFedora Core 4

[Ming Z. <mi...@el...>]

On Fri, 2006-03-17 at 10:20 -0800, Christensen, ReedX K wrote:
> 
> Hi Ming,
> 
> Sorry about the confusing comment in the "iometer.nonz" patch.  I was
> using Iometer to generate workloads for iSCSI Initiators and Targets,
> which compute CRC values.  A member of our team suggested that having
> non-zero data values would affect the performance of CRC calculations by
> the processor, and it did!

o. ic. yes, if u enable the header/data digest. good catch. i test iscis
target a lot while i never think about that. though i never enable
digest any way. ;)

> 
> I guess the point is that for little cost (done once during init of the
> write buffer) we can get "interesting" data to be written by Iometer.

then shall we make this to be a option as well?

0, all zeroed
1, all random.

so u can observe the crc effect. :P

 
> 
> Reed
> 
> -----Original Message-----
> From: Ming Zhang [mailto:mi...@el...] 
> Sent: Thursday, March 16, 2006 6:21 PM
> To: Christensen, ReedX K
> Cc: iom...@li...; rm....@wi...;
> iometer-devel-list
> Subject: Re: Iometer changes (individually packaged) to run 'dynamo'
> onFedora Core 4
> 
> Hi Reed
> 
> Regarding the patch
> 
> * iometer.zero. it was in cvs after last release.
> 
> * iometer.minmax. could u check if this will be ok instead of using that
> patch. -- in Makefile-Linux. remove that "-DNOMINMAX"
> 
> * iometer.odir. yes, i always hesitate about this. whether leave it
> unchanged, or make it O_SYNC, or make it O_DIRECT. maybe most people
> want bypass the file system or cache effects and test the disk/storage
> system performance? maybe we should make it configurable from command
> line? also another issue is that O_DIRECT is quite broken in 2.4.x
> kernel. anyway, i will apply it first. 
> 
> * iometer.noz. u comment is that processor compute crc. i am a little
> confused here. which processor? main processor in host or storage
> processor. i do not see where we computer crc. or u mean crc computation
> in disk?
> 
> * iometer.port. merged.
> 
> * iometer.jiff. i want to use a better way to do this. see
> http://www.i-scream.org/libstatgrab/. 
> 
> * iometer.blks was in cvs since last release.
> 
> * i need to chow the info on iometer.rand a bit more. eyes blurred
> already after whole day work.
> 
> Thanks!
> 
> Ming
> 
>