Skip to Content.
Sympa Menu

perfsonar-user - Re: [perfsonar-user] Different results from iperf3 vs. bwctl -T iperf3

Subject: perfSONAR User Q&A and Other Discussion

List archive

Re: [perfsonar-user] Different results from iperf3 vs. bwctl -T iperf3


Chronological Thread 
  • From: Brian Candler <>
  • To: Aaron Brown <>
  • Cc: "" <>
  • Subject: Re: [perfsonar-user] Different results from iperf3 vs. bwctl -T iperf3
  • Date: Thu, 11 Dec 2014 00:04:28 +0000
  • Domainkey-signature: a=rsa-sha1; c=nofws; d=pobox.com; h=message-id:date :from:mime-version:to:cc:subject:references:in-reply-to :content-type:content-transfer-encoding; q=dns; s=sasl; b=bu0f17 YPjqTmBW3aYei1TuE/5eIhSCxU+d3o0E8novCPcIOY7gIv8rN/UFWTr0FCheSpXo rSpjnED4P0/uXle6ZJmfxvvFoIO1RaslJWYtMO8fSJ7zz2//ZnjrfNyZvObBqf/f QqbY7LG0q88LleA1Oa1DLIif/jSLqpWmPgXBI=

I made little headway in narrowing this down.

I added some debugging in bwctl.c wait_for_results. When the total test times were ~45 seconds, I was able to see that the difference between the req_time and the current time was 35 seconds (but duration=10, omit=0, fuzz=just over 1 second). So it's something to do with calculating the future request time.

I have also seen cases of wait_for_results saying times of over 80 seconds.

I then added further debugging, but by the time I did this, it was back to "working" again - 15 seconds per test :-(

At bwctl client side:

bwctl: BWLGetTimeStamp: errest=2100000
bwctl: BWLSetTimeStampError: val=9019431321
bwctl: BWLSetTimeStampError: scale=26, multiplier=135
bwctl: BWLGetTimeStamp: errest=2100000
bwctl: BWLSetTimeStampError: val=9019431321
bwctl: BWLSetTimeStampError: scale=26, multiplier=135

That's fine: 0.1 is the default skew estimate, and I put -a 2 on the command line.

On bwctld:

bwctld[16642]: FILE=time.c, LINE=395, BWLGetTimeStamp: errest=100000
bwctld[16642]: FILE=time.c, LINE=718, BWLSetTimeStampError: val=429496729
bwctld[16642]: FILE=time.c, LINE=736, BWLSetTimeStampError: scale=21, multiplier=205
bwctld[16642]: FILE=time.c, LINE=395, BWLGetTimeStamp: errest=100000
bwctld[16642]: FILE=time.c, LINE=718, BWLSetTimeStampError: val=429496729
bwctld[16642]: FILE=time.c, LINE=736, BWLSetTimeStampError: scale=21, multiplier=205

This is all fine. 429496729.0/(1<<32) ~= 0.1 seconds.

I can only guess that in some circumstances large fuzz or offset values are being used, but fiddling with BWCTL_DEBUG_TIMEOFFSET I am not able to recreate the problem.

Unfortunately I'm running out of time now as (a) it's late, and (b) tomorrow these boxes get shipped out :-(

Regards,

Brian.




Archive powered by MHonArc 2.6.16.

Top of Page