greater than gigabit throughput observed using iometer on a gigabit ethernet line
|
View:
New views
5 Messages
—
Rating Filter:
Alert me
|
 |
greater than gigabit throughput observed using iometer on a gigabit ethernet line
by Pradeep S-2
::
Rate this Message:
|
|
|
Re: greater than gigabit throughput observed using iometer on a gigabit ethernet line
by Vedran Degoricija
::
Rate this Message:
Some parts of this message have been removed.
Learn more about Nabble's security policy.
IOmeter reports the aggregate throughput as seen by both sides of the network connection. So if effectively doubles the actual number. You can verify this with perfmon if it shows ~65MB/s. In your Ometer's results display, you should select either read or write summary counters instead of the combined value.
Regards,
Ved From: Pradeep S <pradeepsuresh@...> To: iometer-user@... Sent: Monday, February 23, 2009 3:12:32 PM Subject: [Iometer-user] greater than gigabit throughput observed using iometer on a gigabit ethernet line Hello all, I have some interesting results with some Iometer testing I've been trying to do on an Openfiler disk. When I make my transfer request size incredibly big (4 MB +) my observed throughput (be it for a 2 minute run or a 20 minute run, 4 workers, queue depth of 10 per worker, 20 GB disk size, 50% random, 50% read) are around 130 MBps!! While this is physically impossible on a 1 gigabit ethernet line (am using software iSCSI without jumbo frames), since the theoretical max should be a 125 MBps without considering any overheads, this leads me to the question if Iometer is producing artificial results due to some kind of read write caching? write caching is disabled in my windows environment as well, so i really dont know if this is a bug with iometer, although i sincerely hope its not since i use it heavily for my project. Please let me know if anybody has seen similar behavior (physically impossible throughputs) with Iometer. Thank you, Pradeep ------------------------------------------------------------------------------ Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA -OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise -Strategies to boost innovation and cut costs with open source participation -Receive a $600 discount off the registration fee with the source code: SFAD http://p.sf.net/sfu/XcvMzF8H _______________________________________________ Iometer-user mailing list Iometer-user@... https://lists.sourceforge.net/lists/listinfo/iometer-user ------------------------------------------------------------------------------ Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA -OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise -Strategies to boost innovation and cut costs with open source participation -Receive a $600 discount off the registration fee with the source code: SFAD http://p.sf.net/sfu/XcvMzF8H _______________________________________________ Iometer-user mailing list Iometer-user@... https://lists.sourceforge.net/lists/listinfo/iometer-user |
|
|
Re: greater than gigabit throughput observed usingiometer on a gigabit ethernet line
by Mark_Lindholm
::
Rate this Message:
|
|
|
Re: greater than gigabit throughput observed usingiometer on a gigabit ethernet line
by Joe Eiler-2
::
Rate this Message:
Some parts of this message have been removed.
Learn more about Nabble's security policy.
Ved, I don't think he is measuring a network
connection(using network targets), he is doing "disk" accesses to an iSCSI
target.
What Ved says is true when using network targets;
to do network testing Iometer has to have 2 network targets and the default
display shows the aggregate of all managers and workers. In that case, you
may want to drag the appropriate workers to the results display, so you are just
looking at "one side" of the connection.
Pradeep, your ethernet link is running full duplex
(i.e. can do reads and writes at the same time). Since you have a mix of
reads and writes the ethernet link can sustain fullline rate in each
direction.
The best I have seen for an iSCSI target was
180MB/s (100MB/s read / 80 MB/s write), either direction alone could do about
115MB/s.
The other thing you have to be on the lookout for
is cache hits, both on the local system and on storage box.
Good luck with your testing.
Joe
------------------------------------------------------------------------------ Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA -OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise -Strategies to boost innovation and cut costs with open source participation -Receive a $600 discount off the registration fee with the source code: SFAD http://p.sf.net/sfu/XcvMzF8H _______________________________________________ Iometer-user mailing list Iometer-user@... https://lists.sourceforge.net/lists/listinfo/iometer-user |
|
|
Re: greater than gigabit throughput observed usingiometer on a gigabit ethernet line
by Vedran Degoricija
::
Rate this Message:
Some parts of this message have been removed.
Learn more about Nabble's security policy.
Joe, Thanks for the correction, my bad! I saw "gigabit ethernet" and filtered out the rest. :)
Ved
From: Joe Eiler <dev@...> To: Pradeep S <pradeepsuresh@...>; iometer-user@... Sent: Monday, February 23, 2009 4:18:30 PM Subject: Re: [Iometer-user] greater than gigabit throughput observed usingiometer on a gigabit ethernet line Ved, I don't think he is measuring a network connection(using network targets), he is doing "disk" accesses to an iSCSI target.
What Ved says is true when using network targets; to do network testing Iometer has to have 2 network targets and the default display shows the aggregate of all managers and workers. In that case, you may want to drag the appropriate workers to the results display, so you are just looking at "one side" of the connection.
Pradeep, your ethernet link is running full duplex (i.e. can do reads and writes at the same time). Since you have a mix of reads and writes the ethernet link can sustain fullline rate in each direction.
The best I have seen for an iSCSI target was 180MB/s (100MB/s read / 80 MB/s write), either direction alone could do about 115MB/s.
The other thing you have to be on the lookout for is cache hits, both on the local system and on storage box.
Good luck with your testing.
Joe
------------------------------------------------------------------------------ Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA -OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise -Strategies to boost innovation and cut costs with open source participation -Receive a $600 discount off the registration fee with the source code: SFAD http://p.sf.net/sfu/XcvMzF8H _______________________________________________ Iometer-user mailing list Iometer-user@... https://lists.sourceforge.net/lists/listinfo/iometer-user |
| Free embeddable forum powered by Nabble | Forum Help |
