[c-nsp] SAA processing time correction in VRF

[Robert Kiessling]

We have trouble with the precision of SAA in a VRF (MPLS L3 VPN). It
seems the that time correction magic for udpEcho is not applied inside
a VRF.

To show this, let's see some typical values:

Type               min / max / avg / stddev
udpEcho (global):  16  / 19  / 18  /  0.62
udpEcho (VRF):     16  / 334 / 32  / 50.61
echo (global):     15  / 244 / 31  / 49.60
echo (VRF):        15  / 339 / 32  / 50.47

"udpEcho" is udpEcho(5) with protocol ipUdpEchoAppl and "echo" is type
echo(1) with protocol ipIcmpEcho.

"VRF" uses a Loopback interface in a locally defined RFC2547bis VPN
and the appropriate SAA option to perform the action inside the
VRF. "global" uses normal IP.

These values are collected over thousands of tests, over different
platforms and IOSes 12.2(25)S5 and 12.3(10a). They are very
consistent.

It looks like the time correction is not applied to packets entering
and/or leaving as MPLS.

Anyone with similar experiences?
Any idea why or hints where this is explained?

Robert
Date: Tue, 20 Sep 2005 23:40:51 +0200
Message-ID: <ygjoe6nwh7g.fsf at joseba.easynet.de>
User-Agent: Gnus/5.1006 (Gnus v5.10.6) Emacs/21.3 (berkeley-unix)