[j-nsp] sudden jump in ping response time

[Mark Kamichoff]

Hi Simon - 

On Thu, Jul 10, 2008 at 02:36:49PM -0400, CHEN Xu wrote:
> I am using multiple logical routers on a juniper router, they are
> connected through logical tunnels, or VLANs, or L2VPN, etc.
> 
> When I ping from one logical router to another, I always see some
> glitch in ping response time, like this:
> 
> 64 bytes from 1.1.2.1: icmp_seq=153 ttl=64 time=1.015 ms
> 64 bytes from 1.1.2.1: icmp_seq=154 ttl=64 time=1.018 ms
> 64 bytes from 1.1.2.1: icmp_seq=155 ttl=64 time=1.048 ms
> 64 bytes from 1.1.2.1: icmp_seq=156 ttl=64 time=0.988 ms
> 64 bytes from 1.1.2.1: icmp_seq=157 ttl=64 time=36.950 ms  <----
> 64 bytes from 1.1.2.1: icmp_seq=158 ttl=64 time=1.004 ms
> 64 bytes from 1.1.2.1: icmp_seq=159 ttl=64 time=1.005 ms
> 64 bytes from 1.1.2.1: icmp_seq=160 ttl=64 time=1.003 ms
> 
> This almost always happens in all setups that I played with. Does
> anyone know what's going on and how to fix it?

I believe this is par for the course in the JUNOS architecture with any
type of setup.  ICMP echo replies (and other ICMP messages) are given
the absolute lowest priority, so anything in the control plane will take
priority.  I can see this too on an almost completely idle J2320, via
MTR:

 Host                     Loss%  Snt  Last   Avg  Best  Wrst StDev
 1. 2001:4830:122d:7::1    0.0%    6   0.7   0.6   0.5   0.7   0.1
 2. 2001:4830:122d:19::2   0.0%    6   3.6 103.6   3.2 603.6 244.9
 3. 2001:4830:122d:c::2    0.0%    6   4.4   6.3   4.3  12.4   3.4

The 2nd hop is the J2320.

The JUNOS Enterprise Routing book discusses this briefly (pages 519-520)
in the CoS chapter, and even shows a PING example demonstrating the
large variance in response times, too.  It then goes on to discuss how
RPM can perform the ICMP timestamp reply in hardware (or via the RT
thread on the J-series), for gathering more accurate performance
metrics.  Depending on your needs, RPM may be something to consider
testing.

Hope this helps.

- Mark

-- 
Mark Kamichoff
prox at prolixium.com
http://prolixium.com/
Rensselaer Polytechnic Institute, Class of 2004
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: Digital signature
URL: <https://puck.nether.net/pipermail/juniper-nsp/attachments/20080711/e815e508/attachment.bin>