Archived from groups: comp.dcom.lans.ethernet (
More info?)
In article <dc998cfd.0410192104.4d384174@posting.google.com>,
aaabbb16@hotmail.com (wld) wrote:
> Hi,
> I hava a switch Support 802.d layer two based CoS (can set each port
> for 0-7 priorty.
Do you mean that you are assigning *frame* priority based on the arrival
port? Remember, "priority" in 802.1D means the priority assigned to a
*frame*. Assigning frame priority based on arrival port seems odd to me.
It implies that *all* frames arriving on some port are "more important"
than *all* frames arriving on some other port. If the intent is to give
particular *users* higher priority, the assignment should rather be
based on MAC address (not port), so that the priority assignment "moves"
with the station.
Nonetheless, let's continue with your hypothetical situation (perhaps it
is a school assignment?).
> in fact it map to four queues e.g 0-2 ->q1, 3-4
> -->q2, 5-->q3, and 6-7->q4 and it is weight round robin 10:5:2:1
> ->q4:q3:q2:q1)
You have four classes of service on each output port. This is fairly
common.
Is the weighting being performed on a *frame* basis (i.e., 10 frames, 5
frames, 2 frames, 1 frame) or on a *byte* basis (e.g., 100K bytes, 50K
bytes, 20K bytes, 10K bytes)? Either is possible, and the resulting
traffic will vary. For simplicity, I will assume that the weighting is
on a byte basis; if all frames are assumed to be the same length (bad
assumption in practice), there is no difference between the two schemes.
By the way be *very careful* pronouncing "four queue system" in English!
:^)
> The question is that suppose i have two port (p1 and
> p2)using line speed send diff traffic
What do you mean by "diff traffic"? Differentiated service? Different?
> go to p3 and set p1 to 7 and p2
> to 5. and all ports are same line speed without flow control.
>
So your scenario has two steady-state, wire-speed frame streams, one at
priority 7 (highest class of service) and one at priority 5 (next
highest class of service) targeted to a single output port.
> Does p3 drop total half packets and 1/3 from p1 and 2/3 from p2
> (10:5)?
>
Using the weightings you provided, frames arriving from port 1 will
consume 10/15 (two-thirds) of the capacity of port 3, and frames
arriving from port 2 will consume 5/15 (one-third) of the capacity.
Assuming that this situation lasts "forever" (i.e., steady-state), and
the switch has finite buffers, 1/3 of the frames arriving on port 1, and
2/3 of the frames arriving on port 2 will ultimately be discarded.
> if p3 does not congestion, does this CoS still work ?
>
"Work" for what purpose? Obviously, if the purpose is to forward all of
the frames, no class-of-service scheme "works" if the offered load
exceeds the switch or port capacity in the steady-state. I discuss this
problem at length in Chapter 13 of "The Switch Book". Priority provides
a means to ride-through transient overload conditions without adversely
impacting the performance of time-sensitive applications; it cannot
resolve a situation where the steady-state demands placed on either the
switch or a link exceeds the capacity of the switch or link.
>
> if it connect another switch via trunk, does this CoS still work?
>
Same answer as above.
> Does incoming tranffic add tag when it going to source port in the
> switch and take off this tag when it going out of dest. port?
>
I don't understand this question. A tag-aware switch should tag untagged
frames for transmission on a "tagged" output port; similarly it should
strip the tag from received frames when transmitting them onto an
"untagged" output port.
>
> What is diff. between QoS and CoS?
>
That all depends on how you define QoS and CoS; many people use
different (and conflicting) definitions, so you get different answers.
This is also discussed at length in Chapter 13.
--
Rich Seifert Networks and Communications Consulting
21885 Bear Creek Way
(408) 395-5700 Los Gatos, CA 95033
(408) 228-0803 FAX
Send replies to: usenet at richseifert dot com