[j-nsp] Traffic billing - L2 encap to include or not?

[Richard A Steenbergen]

On Sat, Jun 13, 2009 at 12:53:32PM -0500, Richard A Steenbergen wrote:
> > PS2: Oh, maybe someone could check on J switches - would be nice to
> >      know ...
> 
> The MX uses essentially the same type of hardware as a T-series, but
> with an extra EZChip ASIC added to do some basic Ethernet framing. You
> actually pose a good question, it might be possible for them to count
> the L2 overhead in SNMP, but I'm not sure if they actually do it or not. 
> EX is an entirely different architecture from traditional Juniper
> routers so I wouldn't even begin to guess.
> 
> At any rate, this is a discussion better suited for juniper-nsp mailing
> list.

Moved this thread from NANOG to Juniper-nsp... 

I just did some quicky tests over a untagged 10GE between a MX960
running 9.1R2, and a EX8208 running 9.6B2, sending a 1500 byte (ip size)
packet through the devices. Anything above 1500 is a layer 2 header.

On the EX physical interface (for both hw forwarded and local packets):

  Input bytes:                      1518 (12144 bps)                 [1518]
  Output bytes:                     1518 (12144 bps)                 [1518]

On the EX logical interface (hw rx, local re destination)

  Input bytes:                      1500 (0 bps)                     [1500]
  Output bytes:                        0 (0 bps)                        [0]

On the MX physical interface (hw rx, local re generated reply, hw tx)

  Input bytes:                      1500 (3328 bps)                  [1500]
  Output bytes:                     1522 (6088 bps)                  [1522]

On the MX logical interface (same as above)

Remote statistics:
  Input bytes:                      1500 (0 bps)                        [0]
  Output bytes:                        0 (0 bps)                        [0]

On the MX physical interface (hw rx, hw tx, no local re involved)

  Input bytes:                      1500 (3328 bps)                  [1500]
  Output bytes:                     1500 (7032 bps)                  [1500] 

So, based on these quicky tests...

It seems like MX only accounts for L2 headers when the packet is coming
from the RE, not from other hardware interfaces. Also, I'm not sure
where the value of 1522 comes from, but I don't see how it could be
correct. This is an untagged interface, so I would expect to see 1514
(1500 + ethernet header) or 1518 (same + FCS). I'm looking at the packet 
as it leaves the router hw, so I shouldn't be seeing any tnp overhead or 
anything like that.

The EX on the other hand seems to be recording the L2 headers on the
counters for the physical port, though it is unclear if the 1518 value 
is inclusive of the FCS, or if its just forcing a vlan to be virtually 
counted even though there is no tag on this packet.

The logical interfaces of both these platforms seem to be recording only
the post-L2 information. Obviously this was a very quick and dirty test
(it was all I could do in 15 minutes before heading to the airport). It
should probably be re-tested with untagged, vlan-tagging, flexible
vlan-tagging (which reserves space for a 2 vlan stack) combinations, and
better set up to isolate the differences between pure hw forwarded
packets and RE handled traffic. Still, I thought it would be
interesting to folks.

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)