Latency based path selection
Is it possible to make route selection based on path latency? (or even delegate path selection to an external app/program/api?) I'm currently using ospf (but can really use any igp) to manage routing in a system with many end nodes all connected together via a handful of servers (via vpn links) which are spread across the country. The problem is, the 'end nodes' are always moving/relocating and currently have to be manually updated to use which ever server(s) are closest / quickest. It would be nice if there was a method or protocol that could make path selection based on lowest avg latency. The applications that transverse the network aren't bandwidth intensive, but are very latency sensitive so detecting and using the lowest latency path would ideal. Any input / suggestions would be greatly appreciated!
On 28 February 2013 23:06, Ryan Whelan <rcwhelan@gmail.com> wrote:
Is it possible to make route selection based on path latency? (or even delegate path selection to an external app/program/api?) [...]
Any input / suggestions would be greatly appreciated!
This might sound a little too obvious, but why not just set the metric of the link to the number of micro/milliseconds measured across it? M
Le 01.03.2013 00:09, Matthew Walster a écrit :
On 28 February 2013 23:06, Ryan Whelan <rcwhelan@gmail.com> wrote:
Is it possible to make route selection based on path latency? (or even delegate path selection to an external app/program/api?) [...]
Any input / suggestions would be greatly appreciated!
This might sound a little too obvious, but why not just set the metric of the link to the number of micro/milliseconds measured across it?
Because : 1/ that is not an automatic process 2/ you can't do real time adaptation to the latency But unfortunately, this is probably the simplest/easiest way to do it.
On 1 March 2013 08:07, JLP <b2a083e4fcb25ddd@mateu.be> wrote:
1/ that is not an automatic process
It could be, if you regenerated your OSPF config every (for instance) hour and signalled BIRD to reload it's configuration files...
2/ you can't do real time adaptation to the latency
Any kind of loading on the links will affect the latency. Yes, you might have protected circuits that will have variable latency, but generally you'll know about these events and can take actions accordingly. M
On Fri, Mar 01, 2013 at 09:07:56AM +0100, JLP wrote:
Le 01.03.2013 00:09, Matthew Walster a écrit :
On 28 February 2013 23:06, Ryan Whelan <rcwhelan@gmail.com> wrote:
Is it possible to make route selection based on path latency? (or even delegate path selection to an external app/program/api?) [...]
Any input / suggestions would be greatly appreciated!
This might sound a little too obvious, but why not just set the metric of the link to the number of micro/milliseconds measured across it?
Because : 1/ that is not an automatic process 2/ you can't do real time adaptation to the latency
That would cause the routing protocol to flap constantly because as you move to a low latency link the traffic moves as well so it becomes a high latency link, while the previous link will get less traffic therefore it will have a lower latency and opsf would flap again. cisco eigrp has a possibility to include load of the interface in the path selection but cisco themselves don't recommend it's use. also eigrp means an all cisco network. mk
On Thu, Feb 28, 2013 at 11:09:34PM +0000, Matthew Walster wrote:
This might sound a little too obvious, but why not just set the metric of the link to the number of micro/milliseconds measured across it?
Because latency depends on traffic, so that would probably cause permanent oscillations in OSPF. -- Elen sila lumenn' omentielvo Ondrej 'SanTiago' Zajicek (email: santiago@crfreenet.org) OpenPGP encrypted e-mails preferred (KeyID 0x11DEADC3, wwwkeys.pgp.net) "To err is human -- to blame it on a computer is even more so."
participants (5)
-
JLP -
Martin Kraus -
Matthew Walster -
Ondrej Zajicek -
Ryan Whelan