topic Re: sflow, pstag, and order of packet operations. in ExtremeSwitching (EXOS/Switch Engine)

sflow, pstag, and order of packet operations.

Nishal_Goburdha — Thu, 24 Jan 2019 15:39:25 GMT

hi,

(hardware = x670g2; software = 16.1.4.2-patch1-3)

i'm looking for some validation of whether this is a feature, or a bug, please.
we have several ports that have pstag(s) enabled, eg:

configure vlan bob tag 100
configure vlan jane tag 101

configure vlan bob add ports 1 tagged 11
configure vlan jane add ports 1 tagged 12

we export sampled, ingress sflow, on a per port basis.
in our case, the goal is to approximate mac-address to mac-address in/out bytes.

our collector sees the export fine, however, the vlan-id in the export reads, either as 11 or 12, and not bob or jane. for our purposes, we'd prefer that this be in the "destination" vlan (ie. bob / jane).
so i'm guessing that sflow export is done _before_ processing of the packet/frame, since, the "source" vlan appears to be the one listed.

we can work around this with a software hack, but i'd really prefer not to. so, before effecting any such workarounds, i thought i'd ask:
# if this is known and intentional?
# what the order of operations on the packet would be, in the _egress_ direction (ie. if i changed the way i collected sflow export)
# if this is standard across other extreme switching gear (specifically x690, x870, and the newer SLX gear)

pointers to relevant documentation would be really helpful.

Re: sflow, pstag, and order of packet operations.

StephenW — Thu, 24 Jan 2019 22:22:35 GMT

I would expect this to be correct behavior as it captures the raw packet header that is to be sent. Looking at RFC 3176 it doesn't document the scenario you mention.

Re: sflow, pstag, and order of packet operations.

Nishal_Goburdha — Mon, 28 Jan 2019 13:29:11 GMT

hi,

thanks for the answer. i did not mean to imply that this is not correct; i'm curious about implementation - ie. order of packet operations - so that we can adopt and code, a standard approach. is there maybe a reference to this (order of packet operations) somewhere?

our specific need is to be able to see traffic from a set of mac addresses, within a particular vlan (bob or jane) so the c-tag, as we're seeing is, it less than ideal, since there may be _many_ of those.