AF:
NF:0
PS:10
SRH:1
SFN:
DSR:
MID:<20070205103425.0ef7aaa2@ripper.onstor.net>
CFG:
PT:0
S:andy.sharp@onstor.com
RQ:
SSV:onstor-exch02.onstor.net
NSV:
SSH:
R:<brian.deforest@onstor.com>,<raj.kumar@onstor.com>,<dl-designreview@onstor.com>,<dl-qa@onstor.com>
MAID:1
X-Sylpheed-Privacy-System:
X-Sylpheed-Sign:0
SCF:#mh/Mailbox/sent
RMID:#mh/Mailbox/design review	0	BB375AF679D4A34E9CA8DFA650E2B04E0250257E@onstor-exch02.onstor.net
X-Sylpheed-End-Special-Headers: 1
Date: Mon, 5 Feb 2007 10:36:04 -0800
From: Andrew Sharp <andy.sharp@onstor.com>
To: "Brian DeForest" <brian.deforest@onstor.com>
Cc: "Raj Kumar" <raj.kumar@onstor.com>, "dl-Design Review"
 <dl-designreview@onstor.com>, "dl-QA" <dl-qa@onstor.com>
Subject: Re: Test Plan for Small file performance
Message-ID: <20070205103604.378bfcdc@ripper.onstor.net>
In-Reply-To: <BB375AF679D4A34E9CA8DFA650E2B04E0250257E@onstor-exch02.onstor.net>
References: <BB375AF679D4A34E9CA8DFA650E2B04E01D4D471@onstor-exch02.onstor.net>
	<BB375AF679D4A34E9CA8DFA650E2B04E0250257E@onstor-exch02.onstor.net>
Organization: Onstor
X-Mailer: Sylpheed-Claws 2.6.0 (GTK+ 2.8.20; x86_64-pc-linux-gnu)
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit

I wanted to add something that I didn't think of in the review meeting,
and that is that the performance tests should start with an expected
result *before* the tests are run.  For example, the person
implementing the test should determine ahead of time roughly how long
the operation is expected to take if the filer has decent performance.

So for a 'cp -r' test, an expected number could be arrived at by trying
the same test against various Unix NFS servers (Linux, Solaris) with
DAS, and those numbers could be combined with a 'pencil in the ceiling'
calculation, like 'transfering 500MB of files, 25,000 files, should
xfer at roughly 10MB/s, so 50 seconds, give or take'.  The 10MB/s
figure comes from assuming that while some files might be multiple MBs
in size, most are less than 10KB in size.

That way if the results are way off, we'll know right away, and we'll
also know if the [expected] improvement from Lambo to Delo is
improvement enough.

But I'm sure everyone knew that already.  I'm just sayin'.

Cheers,

a 

On Mon, 5 Feb 2007 10:14:35 -0800 "Brian DeForest"
<brian.deforest@onstor.com> wrote:

> Raj, a few suggestions were mentioned in the design review that would
> provide additional data points.
> 
> In addition to 30,000 files per test, you could add another dimension
> to the testing to vary the number of files in the directory, e.g. 100,
> 1000, 2000, ...32000 files.   This would verify performance scales
> linearly (as expected) as well as cover the range of likely/typical
> customer tests..   Cutting/pasting 30,000 files from explorer may
> take a looong time...   (Not sure why this is a magic number).  
> 
> The other area mentioned was to include tar/cpio testing to determine
> if any additional (future) work is needed and also determine if any
> performance improvements are achieved in Delorean (though not a goal
> or requirement).  Rsync and robocopy testing were mentioned for
> similar reasons.
> 
> _____________________________________________
> From: Raj Kumar 
> Sent: Wednesday, December 27, 2006 11:26 AM
> To: dl-Design Review; dl-QA
> Subject: Test Plan for Small file performance
> 
> Hi,
> 
> This is the test plan based on the Delorean PRD & the draft functional
> specification for "small file performance". Please review and send the
> comments by 01/03/2007.
> 
> Test Plan:
> \\mightydog\Program
> Management\Delorean\QE\TestPlans\SmallFilePerformance_TestPlan_Ver0_1.do
> c
> 
> Related Documents:
> PRD: \\mightydog\Program Management\Delorean\Marketing\Delorean
> MRD-PRD-REV1-6.xls
> Draft functional spec: \\mightydog\Program
> Management\Delorean\QE\DRAFT-Smal fileperformance-Spec.doc
> 
> Thanks.
> 
> --kumar :-)
