[alcatel-nsp] Question about Service QoS Policies on 7750SR

Mon Feb 23 18:38:43 EST 2009

You are losing me a bit on what you are actually trying to do.  Is there
an actual design you are working on or are you just trying to get a
better feel for how the Alcatel routers work?

In regards to schedulers it's my understanding that schedulers don't
reserve anything.  They are simple rules to condition traffic.  They
have no meaning in so far as how a port actually conditions traffic.  I
believe "obtained bandwidth" refers to bandwidth defined in the root
scheduler that a sap ingress/egress queue policy references which seems
to be what you are alluding to in your first interpretation.  So if I
configured:

  scheduler-policy "bandwidth" create

                        tier 1

                scheduler "1.5M" create

                    rate 1500 cir 1500

                exit

The "obtained bandwidth" is 1500 based on (I think) the CIR.  Maybe
"obtained bandwidth" is a little misleading because I actually haven't
obtained or reserved anything from a port stand point.  All I have done
is rate or shaped any traffic from a queue that references the scheduler
to 1.5mbs.  

The same scheduler policy can be applied across multiple ports or saps.
This leads me to believe that a scheduler policy or a scheduler has no
relevance to actually port bandwidth allocation.  So there is no
notation of dividing out bandwidth in a scheduler policy from one port
to the next or one sap to the next.  It can only allocate bandwidth to
its children tiers.

In the case of having more then one SAP defined on the same port and you
want all those SAPs to have an equal chance of sending traffic and
ensure that all traffic is prioritized equally across all SAPs then that
is when you have to look at either a multi-service site policy or a port
scheduler as mentioned previously.

I really don't understand what you mean by "multiple tier-1 schedulers
defined on the same SAP".  As you know we associate traffic to a given
scheduler via the SAP ingress and egress policy via the parent command
under the queues themselves, while the SAP itself references the
scheduler policy.  While you can create a sap policy where the queues
referenced different schedulers if the schedulers don't reside under the
same scheduler policy then I don't think the parent command would do
anything.  Maybe I just don't understand your question.

One thing I do get a little fuzzy on is where the arbitrator fits in.  I
think the arbitrator is the final conditioning of traffic before actual
port serialization or transport through the switch fabric in which case
it would be after queue scheduling.  So in mind after we have done all
these fancy schedulers, traffic still goes through the low priority and
high priority pass.

________________________________

From: Kila Hsu [mailto:hsukila at yahoo.com] 
Sent: Monday, February 23, 2009 6:32 AM
To: alcatel-nsp at puck.nether.net; GARCIA DEL RIO Diego; Mark Griffin
Subject: RE: [alcatel-nsp] Question about Service QoS Policies on 7750SR

Mark,

Thanks for your suggestion. I got this book and read through the
queueing and scheduling parts of it. This helps but most of my question
still remains.

In Chapter 10 of the book, the idea of dual arbitrator makes sense but
it would not be so clear when we have H-Qos and have more than one SAP
defined on the same port, or have multiple tier-1 schedulers defined on
the same SAP.

In Chapter 11 of this book, the arbitrator idea is not mentioned
anymore.. It is assumed that the root scheduler already have the
"Obtained bandwidth" but it is not clear to me how the "obtained
bandwidth" is calculated. Does this depend on the CIR/PIR defined on the
scheduler, or simply by dividing the total BW on the port by the number
of tier 1 schedulers, or even other more complex method? I believe the
actual schedulers does not divide the bandwidth first then schedule the
packets, but rather the BW is the effective value of the result of
scheduling. It is very clear to everybody that how the tier one will
select a packet from all its child queues when it is allowed to
transmit, but it is totally black (for me at least) how the actual port
selects which tier 1 scheduler to send next.
>From the previous response from Diego it seems this is again back to the
arbitrator, but if anyone knows any more details it would be very nice,
since I'm not too confident to deploy this without knowing more
details..

thanks,
Kila

--- On Thu, 2/19/09, Mark Griffin <Mark.Griffin at hawaiiantel.com> wrote:

From: Mark Griffin <Mark.Griffin at hawaiiantel.com>
Subject: RE: [alcatel-nsp] Question about Service QoS Policies on 7750SR
To: hsukila at yahoo.com, alcatel-nsp at puck.nether.net, "GARCIA DEL RIO
Diego" <Diego.Garcia_Del_Rio at alcatel-lucent.be>
Date: Thursday, February 19, 2009, 7:00 PM

Kila, do you have Ram's book "Advance QoS for Multi-service IP/MPLS
Networks"?  His book does a good job of explaining the behavior of
traffic conditioning in the hierarchal model.

________________________________

From: alcatel-nsp-bounces at puck.nether.net
[mailto:alcatel-nsp-bounces at puck.nether.net] On Behalf Of Kila Hsu
Sent: Thursday, February 19, 2009 12:44 PM
To: alcatel-nsp at puck.nether.net; GARCIA DEL RIO Diego
Subject: Re: [alcatel-nsp] Question about Service QoS Policies on 7750SR

Thanks a lot, Diego!

I still have several question though. Assume that we don't have a port
scheduler defined, and we do get congestion. As you mentioned, now
queues are transmitted at the physical level using round robin, with
expedited queues being served first.

1. By "queues" do you mean the queue-to-be-transmitted for each SAP? If
there are multiple tier 1 schedulers, or orphaned queues in a SAP, which
one will get transmitted? (or will these queues now become the
candidates for the physical level round-robin?)

2. How are the weights assigned for all these queues? Or is this just an
plain old round-robin with every queue equally weighted?

3. If under each SAP I have hierarchical scheduler, it is possible that
for one SAP, the next packet to be sent comes from an best-effort queue
but the next-next one comes from an expedited queue (although this is a
bad design). In this case, will this SAP have to wait in the physical
level until the expedite queues from other SAPs are transmitted?

I hope I state the problems clearly. Thanks again for all who kindly
answer!

Best Regards,
Kila Hsu

--- On Mon, 2/16/09, GARCIA DEL RIO Diego
<Diego.Garcia_Del_Rio at alcatel-lucent.be> wrote:

From: GARCIA DEL RIO Diego <Diego.Garcia_Del_Rio at alcatel-lucent.be>
Subject: Re: [alcatel-nsp] Question about Service QoS Policies on 7750SR
To: alcatel-nsp at puck.nether.net
Date: Monday, February 16, 2009, 1:15 PM

Hi Hsu,

            If your version of TiMOS allows it, you could run a
"port-parent" scheduler, which allows you to map the parent scheduler of
the queues in the SAPs to a port-level scheduler that will distribute
bandwidth between all it's children according to the configured
prioritied.

            If you don't have the port-schduler function (it appeared in
TiMOS 5.0r4 and onwards), sap-queues and schedulers behave "as expected"
when there is no congestion at the port level, that is, all egressing
traffic is less than the port's egress rate.

            If there is congestion though, and no port-scheduler is
configured, queues are indeed served in round-robin fashion with the
"expedited" queues (in in-profile state) being served exhaustively, then
"best-effort" (in-profile) and finally all out-of-profile queues.

            So.. in short, if using TiMOS 5.0 or newer, take a look at
port-level schedulers ;-)

Cheers!

Diego Garcia del Rio
New Product Introduction - Network Design Engineer 
IP Division
Alcatel-Lucent - Antwerpen
Mobile: +32 4734 30 245
OnNET 2 605 3871
Phone (+32) 3 2403871 
e-mail diego.garcia_del_rio at alcatel-lucent.com 

<<
This message (including any attachments) contains confidential
information intended for a specific individual and purpose, and is
protected by law. If you are not the intended recipient, you should
delete this message. Any disclosure, copying, or distribution of this
message, or the taking of any action based on it, is strictly prohibited
without the prior consent of its author.
>> 

________________________________

From: alcatel-nsp-bounces at puck.nether.net
[mailto:alcatel-nsp-bounces at puck.nether.net] On Behalf Of Kila Hsu
Sent: 16 February 2009 16:46
To: alcatel-nsp at puck.nether.net
Subject: Re: [alcatel-nsp] Question about Service QoS Policies on 7750SR

Hi,
Thanks a lot to Mark for the information about the ways to share a port
among SAPs.
However, my boss seems to like the idea that each SAP can have its own
queues and schedulers though..
So say if I still need to deploy QoS on a per-SAP base, anyone know how
the packets from different SAPs defined on a common port would be
scheduled, if each has its own queue/scheduler?
>From the config guide I do see the "expedite" setting have something to
do with "hardware scheduler", but it is not clear if the setting is used
in cases like this.
Any idea would be appreciated.

thanks,
Kila Hsu

_______________________________________________

alcatel-nsp mailing list

alcatel-nsp at puck.nether.net

https://puck.nether.net/mailman/listinfo/alcatel-nsp

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://puck.nether.net/pipermail/alcatel-nsp/attachments/20090223/8e819a06/attachment-0001.html>