From: HP-France,ex2 ("SANCHEZ-MONGE,ANTONIO)
Date: Wed Apr 21 2004 - 10:09:02 GMT-3
Hi,
I think shaping takes place even if there is no congestion (CIR or an
equivalent has to be respected).
Cheers,
Ato.
-----Original Message-----
From: Richard Dumoulin [mailto:richard.dumoulin@vanco.es]
Sent: miircoles, 21 de abril de 2004 13:02
To: SANCHEZ-MONGE,ANTONIO (HP-France,ex2); 'William Chen';
ccielab@groupstudy.com
Subject: RE: Traffic Shaping Buffer.
I would think of the decision process like this:
forwading process to determine the output interface --> Shaping process to
determine whether shaping is needed (congestion) (else skip) --> software
queue (wfq) if shaping active (else skip) --> Scheduler determines which
queue to serve --> hardware queue --> physical wire
Cheers
--Richard
-----Original Message-----
From: SANCHEZ-MONGE,ANTONIO (HP-France,ex2)
[mailto:antonio.sanchez-monge@hp.com]
Sent: miircoles, 21 de abril de 2004 12:30
To: Richard Dumoulin; 'William Chen'; ccielab@groupstudy.com
Subject: RE: Traffic Shaping Buffer.
Hi,
The path the packet follows is (correct me if I am wrong):
Routing process -----> Shaping engine -----> Scheduler -----> Interface
The "shape max-buffers" and the "queue limit" are independent values.
The first tells you how many packets exceeding the CIR can be buffered by
the GTS engine.
The second tells you how many packets you can queue for this class if there
is packet congestion (this queue is controlled by the scheduler-WFQ).
So the flowchart for a new packet trying to go out of the interface is:
1) The packet gets to the shaping engine.
1.1) If the token bucket Bc has enough tokens, the packet is sent to the
scheduler.
1.2) If the token bucket does not have enough tokens, and the shaping buffer
has less packets than "max-buffers", the packet is buffered, and wait until
there are enough tokens for the transmission of all the queued packets; as
soon as there are enough tokens to transmit it, it is sent to the scheduler.
1.3) If the token bucket does not have enough tokens, and the shaping buffer
has reached "max-buffers", the packet is dropped.
2) If the packet gets to the scheduler.
2.1) If the queue is empty, it is transmitted.
2.1) If the queue depth is less than "queue limit", it is queued.
2.1) If the queue depth has reached "queue limit", it is dropped.
Cheers,
Ato.
-----Original Message-----
From: Richard Dumoulin [mailto:richard.dumoulin@vanco.es]
Sent: miircoles, 21 de abril de 2004 10:57
To: SANCHEZ-MONGE,ANTONIO (HP-France,ex2); 'William Chen';
ccielab@groupstudy.com
Subject: RE: Traffic Shaping Buffer.
Suppose that in the example of William we have a T1. Here the congestion is
created by the traffic shaping configuration when traffic rate reaches 128
Kbps. So queueing is needed.The type of queueing will be wfq as per
configuration so the queue-limit 256 will count. In this case I think the
traffic buffer will never reach its limits. Now suppose that instead of a T1
we have a 64 Kbps leased line. Here traffic shaping will never take place
and congestion will be created by the line speed itself. Again queueing will
be wfq and the queue limit will count. This is how I understand it. Where am
I wrong ?
By the way this is already GTS but configured with the modular cli right ?
Regards
--Richard
-----Original Message-----
From: SANCHEZ-MONGE,ANTONIO (HP-France,ex2)
[mailto:antonio.sanchez-monge@hp.com]
Sent: miircoles, 21 de abril de 2004 10:18
To: 'William Chen'; ccielab@groupstudy.com
Subject: RE: Traffic Shaping Buffer.
Hi William,
Shaping takes place BEFORE. The shaping algorithm buffers all the packets
that exceeded the CIR up to the "shape max-buffers" value. The packets that
are within the CIR are scheduled. If the interface is not congested, the
packets are then sent rightaway. If it is congested, the fair queue
algorithm takes place and the "queue limit" controls how many packets can be
buffered by the scheduler. For example, imagine you want to shape a flow to
64kbps in a T1 interface. The T1 interface can have its buffers completely
free, so "queue limit" will never kick in. If the real flow for that class
is 128kbps, the "shape max-buffers" will control the packets queued by the
shaping engine. On the other hand, if you configure "bandwidth 1544" for an
interface whose real line rate is 64kbps, you shape a class to 256kbps, and
the real traffic flow for that class is 128kbps, the shaping engine will let
all go through, but the scheduler will have to drop at least 50% of the
incoming packets as soon as "queue limit" kicks in. Cheers,
Ato.
-----Original Message-----
From: William Chen [mailto:kwchen@netvigator.com]
Sent: miircoles, 21 de abril de 2004 2:47
To: ccielab@groupstudy.com
Subject: Traffic Shaping Buffer.
Dear all,
I understand that the command "shape max-buffers xxx" is using in a
CB-Shaping to configure the maximum number of packets will be queued in the
shaping buffer. However, what is the difference of the command with
"queue-limit'? For example: policy-map testing
class class-default
shape average 128000
shape max-buffers 2048
fair-queue
queue-limit 256
R5#sh policy-map testing
Policy Map testing
Class class-default
Traffic Shaping
Average Rate Traffic Shaping
CIR 128000 (bps) Max. Buffers Limit 2048 (Packets)
Weighted Fair Queueing
Flow based Fair Queueing Max Threshold 256 (packets)
Then, how many packets are allowed to be buffered? 256 or 2048?
Moreover, is there any command to do the similar in GTS?
Best Regards,
William Chen
This archive was generated by hypermail 2.1.4 : Mon May 03 2004 - 19:48:51 GMT-3