number of hops != performance
We have competitors that are claiming that their network is superior to ours (salesdroids to customers) because they have fewer L3 hops in their network. I see this "fact" pop up in customer questions all the time. I can see that L3 hops adds latency if a network is built on slow (2meg for instance) links, but at gigabit speeds, L3 hops adds microseconds in latency (if you use equipment that forward using hardware-assisted forwarding, but as far as I know there are no routers out there nowadays that doesnt). Does anyone have a nice reference I can point to to once and for all state that just because a customer has 6-8 L3 hops within our network (all at gigabit speeds or higher) that doesnt automatically mean they are getting bad performance or higher latency. Hiding the L3 hops in a MPLS core (or other L2 switching) doesnt mean customers are getting better performance since equipment today forwards just as quickly on L3 as on L2. -- Mikael Abrahamsson email: swmike@swm.pp.se
In a commercial sense hops are seen as bad, points of failure(?) or 'distance from the middle of the internet'?. Who knows Traceroutes aren't great at seeing whats REALLY going on. I suspect if everyone removed all their 'hop hiding' technology traceroutes would be at least 60% longer, the latency would remain the same. Commercial sense doesn't have to make sense... If its what your competitors use to sell service, Hide your hops ;-) G Mikael Abrahamsson wrote:
We have competitors that are claiming that their network is superior to ours (salesdroids to customers) because they have fewer L3 hops in their network. I see this "fact" pop up in customer questions all the time.
I can see that L3 hops adds latency if a network is built on slow (2meg for instance) links, but at gigabit speeds, L3 hops adds microseconds in latency (if you use equipment that forward using hardware-assisted forwarding, but as far as I know there are no routers out there nowadays that doesnt).
Does anyone have a nice reference I can point to to once and for all state that just because a customer has 6-8 L3 hops within our network (all at gigabit speeds or higher) that doesnt automatically mean they are getting bad performance or higher latency.
Hiding the L3 hops in a MPLS core (or other L2 switching) doesnt mean customers are getting better performance since equipment today forwards just as quickly on L3 as on L2.
-- ____________________________________________________ Message scanned for viruses and dangerous content by <http://www.newnet.co.uk/av/> and believed to be clean
On Tue, Nov 05, 2002 at 06:13:37PM +0100, Mikael Abrahamsson wrote:
We have competitors that are claiming that their network is superior to ours (salesdroids to customers) because they have fewer L3 hops in their network. I see this "fact" pop up in customer questions all the time.
I can see that L3 hops adds latency if a network is built on slow (2meg for instance) links, but at gigabit speeds, L3 hops adds microseconds in latency (if you use equipment that forward using hardware-assisted forwarding, but as far as I know there are no routers out there nowadays that doesnt).
Of course L3 forwarding is not by itself "bad" for the packets. However... If you have a network with "excessive" hops (for some definition of excessive), it probably means one or more of the following: A) you have a poor (or at least non-elegant) network design. B) you have more places for things to go wrong in both hardware and software. C) you're busy gratifying your architectural ego instead of designing the simplest thing possible which gives you the necessary performance and reliability. D) you're buying so much unnecessary hardware that you are either not not financially healthy or you're not passing on as much savings as you could be to your customer. Now while I'm sure that you don't fit into that definition of "excessive", I can think of a few people who do, and they try to use that "but more L3 hops are never bad" argument. -- Richard A Steenbergen <ras@e-gerbil.net> http://www.e-gerbil.net/ras PGP Key ID: 0x138EA177 (67 29 D7 BC E8 18 3E DA B2 46 B3 D8 14 36 FE B6)
Of course L3 forwarding is not by itself "bad" for the packets. However... If you have a network with "excessive" hops (for some definition of excessive), it probably means one or more of the following:
A) you have a poor (or at least non-elegant) network design.
If your L3 topology is well aligned with your L1 topology, you usually end up with more hops. The less intermediate gear, like SONET you use but do L3 instead, the more L3 hops you have.
B) you have more places for things to go wrong in both hardware and software.
This is specifically true for the hop-hiders using MPLS or other mostly pointless multihop recursive switching systems.
C) you're busy gratifying your architectural ego instead of designing the simplest thing possible which gives you the necessary performance and reliability.
Hop-hiding is usually going the other way from the simplest thing.
D) you're buying so much unnecessary hardware that you are either not not financially healthy or you're not passing on as much savings as you could be to your customer.
Eliminating n+1 kinds of gear and replacing it with a smaller number of different kind of boxes makes your network simpler and saves nicely on OPEX. Might be somewhat more CAPEX intensive on the start but not by a large margin.
Now while I'm sure that you don't fit into that definition of "excessive", I can think of a few people who do, and they try to use that "but more L3 hops are never bad" argument.
This would translate to that "hops are bad", regardless of the layer. Many people mess with L4 to L7 generating unneccesary hops on application protocols. These usually seem to be the same growd that does other hiding things. Pete
On Tue, 5 Nov 2002, Petri Helenius wrote:
If your L3 topology is well aligned with your L1 topology, you usually end up with more hops. The less intermediate gear, like SONET you use but do L3 instead, the more L3 hops you have.
This is exactly what we do, we run L3 pretty much directly on the fiber with some OEO-repeaters in between, therefore we display much of our infrastructure in a traceroute. We can do a L2 hop instead, that will probably make things less efficient in some cases and will hide the underlying infrastructure, but will make customers happy. I don't like to do silly technical suboptimisations for cosmetical reasons.
B) you have more places for things to go wrong in both hardware and software.
This is specifically true for the hop-hiders using MPLS or other mostly pointless multihop recursive switching systems.
Quite true. I mean, either the equipment does an L2 or an L3 hop, either way it can go wrong. -- Mikael Abrahamsson email: swmike@swm.pp.se
participants (4)
-
Gary Coates
-
Mikael Abrahamsson
-
Petri Helenius
-
Richard A Steenbergen