Pat [MSFT]
Guest
|
 Posted:
Mon Oct 31, 2005 5:16 pm Post subject:
Re: server/storage load test |
|
|
You would think it would be fairly straight forward - but it isn't.
Virtually all of the stress tools are really designed to stress the
underlying HW - and in fact almost all of them specify WRITE_THROUGH (i.e.
bypass the file cache) for their operations.
So, you can use something like IOMeter to stress the HW and get a feel for
the relative Read & Write perf. It can also give you throughput and latency
info. Keep in mind that IOMeter (like many stress tools) create 1 large
file and basically randomly reads/writes to it. This is very, very
different from creating 100's/1000's of files and opening/closing them.
This is more like a DB Stress environment vs an Office User environment.
Some applications (e.g. Exchange) have storage models already determined
(for Exchange budget .5 IOPS per user during peak usage). So, if you run
IOMeter to determine your storage max IOPS and divide by .5, you will get
the number of Exchange users that it can use. SQL Server has stress tools
that profile the storage based on SQL workloads.
To generate the client count it is really dependent on what types of clients
you are talking about. A lot of stress tools will extrapolate the count but
this should be taken with a large grain of salt. I have seen documentation
where extrapolation estimates an Open/Read/Close operation set to occur
every 15 seconds. This is an extreme assumption for a general file server.
Most users open several files, leave them open for a long time, then close
them.
So, my recommendation would be to model the users. Users that are doing
multi-media editing (as an example) will have a very large Bandwidth
requirement and total storage, but will generate IO's relatively rarely. A
LOB app acting as a user (i.e. one that leverages flat files) may generate a
tremendous amount of IO's, but use only 1 connection resource. Users that
primarily work with Office will have a usage model that is in-between. One
connection will open several files - but they will only cache in only the
views that they need.
On top of all of that the client OS will make a (potentially huge)
difference as the redirector will cache operations as well.
Pat
"Tal Bar-Or" <tal_baror@hotmail.com> wrote in message
news:eDa8qtf3FHA.3036@TK2MSFTNGP15.phx.gbl...
| Quote: | Hello,
I would to formulate method to measure server (file server ,storage )
throughput ,bandwidth during client load.
Finally to get to conclusion what is the X number of clients can server or
storage support and what bandwidth per client is distributed is it linear
way or first in get the most of the bandwidth.
my question is what is the best approach and tools is recommended for such
test?.
Thanks
|
|
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in