MPLS-OPS - Discussion of Multiprotocol Label Switching ()

Robert McCallum | 9 Mar 2007 12:55

RE: : MPLS Convergence Upon Fiber Link Down

MPLS does not increase convergence by default.  Also neither does using BFD.

Robert McCallum 
Senior Consultant

> -----Original Message-----
> From: Peter Macaulay [mailto:peter.macaulay <at> ZDSL.com]
> Sent: 19 March 2007 04:22
> To: mpls-ops <at> mplsrc.com
> Subject: [MPLS-OPS]: MPLS Convergence Upon Fiber Link Down
> 
> I have a network now of 12 nodes with OSPF and BFD (Bidirectional
> Forwarding Detection) with 100 ms timers.  No reply to three BDF
> packets is defined as dead. Each node has 4 GigE fibers. On this
> size network we get convergence in about 1000 ms.
> 
> Our network is about to double to 24 nodes, again with 4 GigE
> fibers.
> 
> I'm thinking the BDF "keep alives" will start to get in the way
> of our user traffic. At what point does a switch to MPLS make sense
> if we want to maintain < 1 second fiber out or node failure convergence?
> Now? Or >20 nodes?  Any help with the math (arithmetic?) would be
> great. Thank you.
> 
> /PETER MACAULAY
>   ZDSL.COM
> 
> -------
> The MPLS-OPS Mailing List

(Continue reading)

headers

VKR V | 9 Mar 2007 16:13

: MPLS load sharing

Hello All,

I know that OSPF can do load sharing to maximum of 6 paths. My question - is load sharing possible based on MPLS label?

Regards

Vic K

Don't get soaked. Take a quick peek at the forecast
with theYahoo! Search weather shortcut.

headers

sthaug | 9 Mar 2007 17:24

Re: : MPLS load sharing

>   I know that OSPF can do load sharing to maximum of 6 paths. My question - is load sharing possible based on
MPLS label?

I sense some misunderstanding here. First, there is nothing specific
in OSPF which says load sharing to a maximum of 6 paths. Second, it is
the *forwarding* part of the router which does the load sharing - OSPF
simply computes routes based on received information.

Different router platforms have different limitations on the number
of paths that can be used for load sharing. The load sharing is often
done by computing a hash of L3 and L4 information (and for switches,
L2 information like MAC address) and using this to ensure that all the
packets in one *flow* (identified by the same L3/L4 information) is
sent along the same path, to prevent reordering of packets.

Load sharing based on MPLS labels simply means that the router can use
MPLS labels as path of the hash. This is fairly common - for instance
the following for Juniper M/T series routers:

http://www.juniper.net/techpubs/software/junos/junos81/swconfig81-policy/html/policy-actions-config12.html#1020348

Steinar Haug, Nethelp consulting, sthaug <at> nethelp.no

headers

Aamer Akhter (aakhter | 9 Mar 2007 21:11

RE: : MPLS load sharing

inline

-- 
Aamer Akhter / aa <at> cisco.com
Ent & Commercial Systems, cisco Systems

> -----Original Message-----
> From: sthaug <at> nethelp.no [mailto:sthaug <at> nethelp.no]
> Sent: Friday, March 09, 2007 11:25 AM
> To: rvkr2005 <at> yahoo.com
> Cc: mpls-ops <at> mplsrc.com
> Subject: Re: [MPLS-OPS]: MPLS load sharing
> 
> >   I know that OSPF can do load sharing to maximum of 6 paths. My
> question - is load sharing possible based on MPLS label?

[AA] 
It is possible, but in reality (depending on implementation) the loadsharing would be done on inner mpls
label (or stack) if the router is unable to determine the IP flow. Also keep in mind it depends on how you use
MPLS (IGP, hop-by-hop, TE, etc). In the some cases you will see no difference than what you had in the
regular IP hop-by-hop behavior, as the FECs are still directed in the same fashion.

See 
http://www1.tools.ietf.org/html/draft-ietf-mpls-ecmp-bcp-03

for some further explanation.

[AA] 

> 
> I sense some misunderstanding here. First, there is nothing specific
> in OSPF which says load sharing to a maximum of 6 paths. Second, it is
> the *forwarding* part of the router which does the load sharing - OSPF
> simply computes routes based on received information.
> 
> Different router platforms have different limitations on the number
> of paths that can be used for load sharing. The load sharing is often
> done by computing a hash of L3 and L4 information (and for switches,
> L2 information like MAC address) and using this to ensure that all the
> packets in one *flow* (identified by the same L3/L4 information) is
> sent along the same path, to prevent reordering of packets.
> 
> Load sharing based on MPLS labels simply means that the router can use
> MPLS labels as path of the hash. This is fairly common - for instance
> the following for Juniper M/T series routers:
> 
> http://www.juniper.net/techpubs/software/junos/junos81/swconfig81-
> policy/html/policy-actions-config12.html#1020348
> 
> Steinar Haug, Nethelp consulting, sthaug <at> nethelp.no
> 
> -------
> The MPLS-OPS Mailing List
> Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
> Archive: http://www.mplsrc.com/mpls-ops_archive.shtml

headers

VKR V | 10 Mar 2007 23:30

RE: : MPLS load sharing

Couples of questions -

1. Aamer, you have mentioned that if router is unable to find determine IP flow (I guess you are talking about IP address in packet) then the router uses inner MPLS label for load sharing. The question is if there is only one label then?

2. What is the sequence of packet flow. My understanding is - PE looks the destination address in the routing table, assigns label to the packet and sends on path-1. The second packet is again looked into the routing table, assign label and sends it on path-2. (assumption is per packet load sharing configured on router).

Is above point 2 correct? If yes then it is not load sharing based on MPLS label but load saring by OSPF.

My question is based on fact that LDP signaling protocol is used and no RS VP configured.

Regards

Vic K.

"Aamer Akhter (aakhter)" <aakhter <at> cisco.com> wrote:

inline

--
Aamer Akhter / aa <at> cisco.com
Ent & Commercial Systems, cisco Systems

> -----Original Message-----
> From: sthaug <at> nethelp.no [mailto:sthaug <at> nethelp.no]
> Sent: Friday, March 09, 2007 11:25 AM
> To: rvkr2005 <at> yahoo.com
> Cc: mpls-ops <at> mplsrc.com
> Subject: Re: [MPLS-OPS]: MPLS load sharing
>
> > I know that OSPF can do load sharing to maximum of 6 paths. My
> question - is load sharing possible based on MPLS label?

[AA]
It is possible, but in reality (depending on implementation) the loadsharing would be done on inner mpls label (or stack) if the router is unable to determine the IP f low. Also keep in mind it depends on how you use MPLS (IGP, hop-by-hop, TE, etc). In the some cases you will see no difference than what you had in the regular IP hop-by-hop behavior, as the FECs are still directed in the same fashion.

See
http://www1.tools.ietf.org/html/draft-ietf-mpls-ecmp-bcp-03

for some further explanation.

[AA]

>
> I sense some misunderstanding here. First, there is nothing specific
> in OSPF which says load sharing to a maximum of 6 paths. Second, it is
> the *forwarding* part of the router which does the load sharing - OSPF
> simply computes routes based on received information.
>
> Different router platforms have different limitations on the number
> of paths that can be used for load sharing. The load sharing is often
> done by computing a hash of L3 and L4 information (and for switches,
> L2 information like MAC address) and using this to ensure that al l the
> packets in one *flow* (identified by the same L3/L4 information) is
> sent along the same path, to prevent reordering of packets.
>
> Load sharing based on MPLS labels simply means that the router can use
> MPLS labels as path of the hash. This is fairly common - for instance
> the following for Juniper M/T series routers:
>
> http://www.juniper.net/techpubs/software/junos/junos81/swconfig81-
> policy/html/policy-actions-config12.html#1020348
>
> Steinar Haug, Nethelp consulting, sthaug <at> nethelp.no
>
> -------
> The MPLS-OPS Mailing List
> Subscribe/Unsubscribe: http://www.mplsrc.com/mplsops.shtml
> Archive: http://www.mplsrc.com/mpls-ops_archive.shtml

Looking for earth-friendly autos?
Browse Top Cars by "Green Rating" at Yahoo! Autos' Green Center.

headers

sthaug | 11 Mar 2007 14:12

Re: : MPLS load sharing

>   2. What is the sequence of packet flow. My understanding is - PE looks the destination address in the
routing table, assigns label to the packet and sends on path-1. The second packet is again looked into the
routing table, assign label and sends it on path-2. (assumption is per packet load sharing configured on router).

Per packet load sharing is a pretty poor assumption these days. Many
modern routers *can not* do per packet load sharing, only per flow.

In any case - the ingress PE will obviously look at the destination IP
address, and can perform load sharing based on this. Other (P) routers
on the path to the destination may or may not be able to look at the IP
address in order to perform load sharing.

>   Is above point 2 correct? If yes then it is not load sharing based on MPLS label but load saring by OSPF.
>    
>   My question is based on fact that LDP signaling protocol is used and no RSVP configured.

LDP or RSVP is irrelevant for the load sharing. OSPF is really also
pretty much irrelevant - there is no "load sharing by OSPF". If the
forwarding component of the router lets the routing component install N
equal cost paths to the same destination, it *does not matter* whether
these paths have been computed by some routing protocol, or for instance
are simply static routes. It is still the forwarding component that does
the actual load sharing, by looking at appropriate parts of the packet.

Steinar Haug, Nethelp consulting, sthaug <at> nethelp.no

headers

Robert McCallum | 9 Mar 2007 16:53

RE: : MPLS load sharing

yes

Robert McCallum
Senior Consultant

  _____  

From: VKR V [mailto:rvkr2005 <at> yahoo.com] 
Sent: 09 March 2007 15:14
To: mpls-ops <at> mplsrc.com
Subject: [MPLS-OPS]: MPLS load sharing

Hello All,

I know that OSPF can do load sharing to maximum of 6 paths. My question - is
load sharing possible based on MPLS label?

Regards

Vic K

  _____  

Don't get soaked. Take a quick peek at the forecast 
with theYahoo! Search weather shortcut.

Note:The information contained in this message may be privileged and confidential and protected from
disclosure . If the reader of this message is not the
intended recipient, or an employee or agent responsible for delivering this message to the intended
recipient, you are hereby notified that any 
dissemination, distribution or copying of this communication is strictly prohibited. If you have
received this communication in error, please notify us
immediately by replying to the message and deleting it from your computer. Thankyou. ThruPoint Ltd.