Re: EMC IP4700 vs NetApp F740 - toasters

10 May 2001


      Barry,
I just re-read my message, and I meant to say that the performance figures
for the EMC were pretty dreadful in your experiment. Sorry about the
confusion.
Regarding EMC's response with running multiple threads.  That seems to
imply that they are running a significant amount of overhead/operation, and
that their system needs to run parallel processes internally to overcome
that overhead.
To apply this statement to our situation, it sounds like the bandwidth to
any one machine may be limited to 40-50Mbits/s read and 20-25Mbits/s write.
However, when multiple machines are using the network, the EMC box will
probably scale quite nicely.
Chris Van Genderen
Software Engineer & Applications Manager
NexFlash Technologies, Inc.
chris_van_genderen@nexflash.com
408-969-4714
Barry Lustig                                                                                            
                    barry@lustig.com            To:     Chris_Van_Genderen@NexFlash.com                                   
                    Sent by:                      cc:     toasters@mathworks.com, (bcc: Chris Van Genderen/Santa            
                    barry@proxy.internal.i        Clara/ISSIHQ)                                                             
                    ssiusa.com                    Subject:     Re: EMC IP4700 vs NetApp F740
05/10/01 12:02 PM
Chris_Van_Genderen@NexFlash.com wrote:
...
When did you perform your evaluation of the IP4700?
Last week.
...
On the F85 performance figures, the performance you are quoting seems
pretty bad.  Is there an unusually small block size in your application,
or
...
another reason you can account for such a difference?
95Mbits/sec throughput on a 100Mbits/sec connection is extremely good.
It is running near the max rate of the connection.  The 95Mbits was the
data rate.  That excludes the packet's header overhead.  I'll have to
check with the engineer who did the actual testing to find out what
block sizes he was using.
...
What was EMC's response regarding performance when you showed them your
data?
They said that we should run multiple threads.  The aggregate bandwidth
abilities of the box could keep the links saturated, but the single
threaded reads and writes ran into the limitations that I've mentioned.
In our application, we need for the single thread to be able to read and
write and high speed.  We have a single data stream coming in off of a
device that has to be written to the filer.
barry
...
Chris Van Genderen
Software Engineer & Applications Manager
NexFlash Technologies, Inc.
chris_van_genderen@nexflash.com
408-969-4714