Bin-

The use of InfTransDelay in OSPF is similar to the TTL in the IPv4
header. Both have seeming "time" semantics, but really operate as maximum
hop counts to remove looping packets (in the case of IPv4's TTL, LSAs in the
case of OSPF) from the network. It doesn't matter that the minimum
InfTransDelay
is much larger than the real transmission delay - the age of LSA can
disperse by as much as MaxAgeDiff (15 minutes) before the routers start
considering them to be separate instances. The router does *not*
delay flooding of LSAs by the configured InfTransDelay value, but always
floods LSAs immediately.

John

-----Original Message-----
From: Bin Liu [mailto:benbin2001HOTMAIL.COM]
Sent: Tuesday, August 07, 2001 6:33 PM
To: OSPFDISCUSS.MICROSOFT.COM
Subject: A student's question 2 (IntTranDelay) and more

Hi everyone,

Thanks for people's help in my question 1 and 3. I am still looking forward
to
getting your help in my question about IntTranDelay. I rewrite the question
here to try to make it clearer :)

In RFC1131, the sample value of the IntTranDelay is 1 second. Could anyone
tell me how this sample value is computed/defined. I think that, in a high
speed network, a time delay in the order of second would cost too much since
such delay happens in every hop of LSA reliable flooding .

Using an example to explain my opinion :) Router a and b are in the same
OSPF routing area. Assuming they are H hops away. Consequently, any LSA
from b to a will take at least H seconds according to the sample
InfTransDelay value. If a is transmitting some files at the speed of 50M
bps to b when b starts to flood its "DOWN_in_1_second" state information
throughout the area, router a will not stop its tranmission until H seconds
later
(at least) . It means that there are (H-1)*50 Mb lost.

In Jim Gray's "the cost of messages" (1988 ACM), the delay of sending or
receiving a message is in the order of millisecond. So my queston is why the
sample value of IntTranDelay in OSPF is set to be 1 second?

My guess is that, since each Link State Update packet can only carry the
LSAs one hop further away from their original point, each intermediate
router
has to prepare the new Link State Update packet . This packet preparing time
dominates the IntTranDelay (in each hop) of the OSPF reliable flooding. If
my
guess is true, then there are two consequent questions: what is the benefit
for
OSPF to only allow the LSU packets carry LSAs one hop away? How to
prove this benefit is large enough to cover the pay of the additional delay
(i.e.,
the packet preparing time in each intermediate router)?

A good dynamic routing scheme should be able to make routers adapt
themselves to environment changes very quickly; and this can only be
achieved by a short end-to-end LSA (i.e., routing state) transmission delay.

Also, where can I get the experiment statistics for the time delays in OSPF
(or any other routing protocol)? Such as the value of IntTranDelay in real
network, how much is the acceptable limitation for the end-to-end LSA
transmission delay, or what is the order of the time for preparing a new LSA
packet.

That's it. Sorry for so many questions in this email :) Thanks for
everybody's
time and thank you for your help in advance.

All the best

yours
Bin

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]