I have a new T3 thats 65msec long. I'd usually be using iperf to test new links, but at 65msec, even at the maximum window size, I can only get 6-8mbit through. No combination of options I've been able to find has gotten me more than 6mbit through this link. Should I just shotgun 9 copies of it? Are there better ways to test these links? Can one verify this link with just a pair of 7200s and linux machines on either side? Or is this something one really needs "real" test hardware for? If 6mbit go through clean, is there a real chance 44 will not? TIA, -mKaegler -- Michael Kaegler, TESSCO Technologies: Engineering, 410 229 1295 Your wireless success, nothing less. http://www.tessco.com/
set your system send and receive TCP buffers larger. You're probably being limited by that. With linux, make sure you have window auto- scaling enabled and have increased the maximum size it can grow to to at least 4MB. Or test with UDP and blast as fast as you can so that you're not seeing TCP weirdness. IOS includes a 'ttcp' command you can use on the router itself: http://www.cisco.com/en/US/tech/tk801/tk36/technologies_tech_note09186a00800... -Dave On May 6, 2009, at 11:10 AM, Kaegler, Mike wrote:
I have a new T3 thats 65msec long. I'd usually be using iperf to test new links, but at 65msec, even at the maximum window size, I can only get 6-8mbit through. No combination of options I've been able to find has gotten me more than 6mbit through this link. Should I just shotgun 9 copies of it?
Are there better ways to test these links? Can one verify this link with just a pair of 7200s and linux machines on either side? Or is this something one really needs "real" test hardware for? If 6mbit go through clean, is there a real chance 44 will not?
TIA, -mKaegler
-- Michael Kaegler, TESSCO Technologies: Engineering, 410 229 1295 Your wireless success, nothing less. http://www.tessco.com/
On Wed, 06 May 2009 11:17:09 -0400, David Andersen <dga@cs.cmu.edu> wrote:
Or test with UDP and blast as fast as you can so that you're not seeing TCP weirdness.
That's the best option... spew packets. Just make sure they are as large as possible without needing fragmentation. And if your setup can support it, set the ethernet MTUs to match the serial interface (4470). Not that PPS should be an issue -- even an ancient NPE150 should be able to process switch enough packets to flood a T3.
Date: Wed, 06 May 2009 11:10:45 -0400 From: "Kaegler, Mike" <KaeglerM@tessco.com>
I have a new T3 thats 65msec long. I'd usually be using iperf to test new links, but at 65msec, even at the maximum window size, I can only get 6-8mbit through. No combination of options I've been able to find has gotten me more than 6mbit through this link. Should I just shotgun 9 copies of it?
Are there better ways to test these links? Can one verify this link with just a pair of 7200s and linux machines on either side? Or is this something one really needs "real" test hardware for? If 6mbit go through clean, is there a real chance 44 will not?
What is "maximum window size"? For T3 at 65 ms (is that one-way or RTT?) you don't need a really big window. Sounds like the system may need tuning. See http://fasterdata.es.net/ I use iperf at multiple Gigabit speeds and it works on a tuned system. You might want to use UDP for testing. It does not care about RTT, but is less efficient to receive. At 45 Mbps, there should be no problems, though. If all else fails, run 4 or 5 iperf jobs in parallel. (Use a different port for each.) -- R. Kevin Oberman, Network Engineer Energy Sciences Network (ESnet) Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab) E-mail: oberman@es.net Phone: +1 510 486-8634 Key fingerprint:059B 2DDF 031C 9BA3 14A4 EADA 927D EBB3 987B 3751
Thanks to everyone who responded on and off-list! It seems evident that I didn't have a complete understanding of the iperf switches which alter buffer sizes. Several people made a few neat points, which I'll quickly summarize: * In iperf, -P will allow one to run multiple tcp tests at once. * IOS has a built-in tester... ttcp. http://tinyurl.com/6fp75j * For suggestions on changing the kernel buffer sizes: * . http://www.29west.com/docs/THPM/ (section 8) * . http://fasterdata.es.net/ * Linux.com has a related writeup: http://www.linux.com/feature/144532 Finally, "past performance does not indicate future results" applies here. 44mbit will not necessarily go clean just because 6 did. Thanks for the tips, -mKaegler On 5/6/09 11:10 AM, "Michael Kaegler" <kaeglerm@tessco.com> wrote:
I have a new T3 thats 65msec long. I'd usually be using iperf to test new links, but at 65msec, even at the maximum window size, I can only get 6-8mbit through. No combination of options I've been able to find has gotten me more than 6mbit through this link. Should I just shotgun 9 copies of it?
Are there better ways to test these links? Can one verify this link with just a pair of 7200s and linux machines on either side? Or is this something one really needs "real" test hardware for? If 6mbit go through clean, is there a real chance 44 will not?
TIA, -mKaegler
-- Michael Kaegler, TESSCO Technologies: Engineering, 410 229 1295 Your wireless success, nothing less. http://www.tessco.com/
participants (4)
-
David Andersen -
Kaegler, Mike -
Kevin Oberman -
Ricky Beam