Re: a question about SE sytle reservation

Terry

Did You got an answer ?

If yes : Pls forward it to me .

If no : as far as i understand

> >                         |
> >           Sends         |       Reserves             Receives
> >                         |
> >                         |       ________
> >    SE( S1{?B} ) <- (a) | (c) |(S1,S2) |   (c) <- SE( (S1,S2){2B} )
> >   &n! bsp;                     |      | {2B} |
> >                         |      |________|
> >    ---------------------|---------------------------------------------
> >                         |
> > SE( (S2){2B} ) <- (b) |
> >                         |
> >                     &nbs! p;   |
> >
> >           Figure 1: Shared-Explicit (SE) Reservation Example

Let us call the above node "R" ,

Let us call the node upstream through interface "a" R1 .

R forward {2B} to R1 but do not make any reservation over interface

"a" .

R1 recieving a request for {2B} over interface "a" .

R1 calculate the Path_TE ( the sum of all Tspecs that were supplied in path

msg's from different previous hob . pls refer to rfc 2205 sec 2.2 page 21) ,

Since their is only one path msg which is from S1 forwarded to "a" so

PATh_Te is equal to Tspec of S1 path msg ie: equal to {1B} .

RSVP process on R1 will issue a "make reservation " call to "Traffic control" process (pls refer to rfc 2205 sec 3.11.2 page 69) .

the parametrs of such call :

Interface which is equal to "a".

TC_flowspec which is equal to {2B}

TC_TSPEC which is Path_Te ie: {1B}

For the traffic control at node R1, will make a reservation over interface "a" which is the min (TC_flowspec,TC_TSpec ) ie: the reservation over interface "a" will be just {1B} .

reg Your enq :

Q1. For this question, we have concern about why the flowspec forwarded upstream to S1 should be {2B} ? The flowspec forwarded upstream to S1 be {1B} can also satisfy the request of Sender S1 .

A1. The flowspec forwarded by R is {2B} because the merging rule do not

take into account any Tspec within path msg(s) .

It is up to R1 to decide (not to R) how much reservation to allocate

over interface "a" .

Q2. Another question. after mergeing , the reservation in S1 should be {2B}, when delete the path created by sender S2, the reservations in S1 also should be {2B}. At that time, S2 doesn't share reservations with S1, S1 only need {1B}, allocate S1 {2B} may be waste. What do you think about this instance ?

A2. No , the reservation at R1 (or S1 if U assume that S1 is directly connected to R ) is not {2B} it is just {1B} as explained above .

When S2 send a PathTear : R modify the RSB (Reservation State Block,

this change will not triger an immediate Resv refresh msg over interface "a" . pls refer to rfc 2205 sec 3.1.5 and rfc 2209 page 9 ) . Also R will initiate a call to the "traffic control" so the result of such call that {1B} is reserved over interface "c" . the Path_tear msg is forwarded over interface "c" toward the downstream node say host H1 .

On the next refresh time , R forward to R1 (over interface "a") a Resv

msg with flowspec {1B} .

From the above , their is no waste reservation resource .

N.B : intially when i read U msg i had the same opinion but the reply from

Lixia Zhang ( one of the author of rfc 2205 ) made me to re-thinck .

Brgds

From: Terry Lee <terrylee@huawei.com>
Date: Thu, 17 Oct 2002 13:36:15 +0800
Cc: mpls-ops@mplsrc.com
Resent-Date: Thu, 17 Oct 2002 02:39:03 -0400
To: Lixia Zhang <lixia@CS.UCLA.EDU>

     For this question, we have concern about why  the  flowspec forwarded upstream 
      to  S1 should be  {2B} ? The  flowspec forwarded upstream to  S1  be  {1B} 
      can  also satisfy the request of Sender S1 . 

     Another question. after mergeing , the reservation in S1 should be {2B},
      when delete the path created by sender S2, the reservations in S1 also should 
      be {2B}. At that time, S2 doesn't share reservations  with S1, S1 only need {1B},
      allocate S1 {2B} may be waste. What do you think about this  instance ?  

----- Original Message ----- 
From: "Lixia Zhang" <lixia@CS.UCLA.EDU>
To: "Terry Lee" <terrylee@huawei.com>; <Braden@isi.edu>; <mpls-ops@mplssrc.com>
Sent: Thursday, October 17, 2002 11:02 AM
Subject: Re: a question about SE sytle reservation


> On 10/16/02 19:54, "Terry Lee" <terrylee@huawei.com> wrote:
> 
> > 
> >                         |
> >           Sends         |       Reserves             Receives
> >                         |
> >                         |       ________
> >    SE( S1{?B} ) <- (a)  |  (c) |(S1,S2) |   (c) <- SE( (S1,S2){2B} )
> >                         |      |  {2B}  |
> >                         |      |________|
> >    ---------------------|---------------------------------------------
> >                         |
> >  SE( (S2){2B} ) <- (b)  |
> >                         |
> >                         |
> > 
> >           Figure 1: Shared-Explicit (SE) Reservation Example
> > 
> > 
> >     Figure 1 shows an example of Shared-Explicit (SE) style
> >     reservations.  There are two upstream senders; packets from sender S1
> >     arrive through previous hop (a), packets from sender S2 arrive through
> >     previous hop (b).
> >     There are only  one downstream receiver R1; packets bound for R1 are
> >     routed via outgoing interface (c) .
> >     Sender S1 request SENDER_TSPEC {1B}, and Sender S2 request SENDER_TSPEC
> > {2B}.
> > 
> >     When SE-style reservations are merged,  the  flowspec forwarded upstream
> >     to  S1 should be {1B} or {2B}?
> 
> 2B (Because it is a shared reservation, one must use the bigger one to fit
> both)
>

The MPLS-OPS Archive

Re: a question about SE sytle reservation