ML wrote:
I'm about to turn on "mls qos" for the first time on a 6509E.
I would like some background information from the QoS experts on this list.
Last time I turned on "mls qos" it was a 3560 which has certain
undesirable defaults when "mls qos" is turned on. I want avoid the same
result with the 6509 which is our Internet edge device. What I want to
accomplish is to mark all incoming traffic from our transit link to CS0.
I don't want to inadvertently get clobbered by a default limit of x% for
egress queue bandwidth that I'm not expecting.
If I understand what I've found out so far:
On the WS-X6724-SFP:
Seems all possible CoS values are mapped to queue 1 for ingress and
egress. The WRR queue ratios are 100,0,0 for queues 1,2,3 (4 is
priority?) So Queue can utilize 100% of the interface bandwidth. So by
default I shouldn't seem traffic getting bottlenecked where it wasn't
before because of some default config?
Well... it depends.
Remember that enabling QoS immediately divides up the transmit and
receive buffer RAM; if you don't map traffic into a queue, the defaults
will mean you're "wasting" (or losing) buffer space, which may or may
not matter depending on your traffic levels.
Recall that there are also the thresholds (the "t" in 2q8t receive or
1p3q8t transmit). Whilst all the CoS values may be mapped to the same
queue, they may not be mapped to the same thresholds, and if the queue
goes above a threshold, WRED or drop may start occuring for one CoS
value but not another.
It really depends on your traffic levels. If you're even close to
filling any links, you want to be very careful about just running with
the defaults. If you've got plenty of headroom, it should be fine.
_______________________________________________
cisco-nsp mailing list [email protected]
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/
