Page 1 of 1

Mpls performance on CCR1036-8G-2S.

Posted: Wed Aug 05, 2020 3:17 am
by leostereo
Hi guys.
Im running a service backbone wich is composed by 3 CCR1036-8G-2S routers.
Triangle topology is formed among them.
8 Gpbs total traffic is running on the backbone.
Im considering implementing mpls in order to manage redundancy and load sharing in a better way.
Question is ...
What about router performance ?
CPU goes at 50% when reaching max traffic. Should I expect higher o lower load after implement mpls?
Is this router suited for mpls ? should I adquiere other model for that ? whitch one ?
Any idea / feedback about this would be wellcome.
Regards.
Leandro.






We receive traffic from two bgp peers.

Re: Mpls performance on CCR1036-8G-2S.

Posted: Wed Aug 05, 2020 3:28 am
by nz_monkey
Hi Leandro,

There are some _major_ issues with MPLS in RouterOS v6. The major ones being the lack of Fast Re-Route, RSVP paths not failing over(or back), NLRI not being updated for L3VPN's and stale labels causing traffic to disappear.

It is usable for basic stuff like VPLS, but it is all done 100% in slow path (no fast path), you would be better off in this situation to just use EoIP tunnels on a OSPF routed L3 backbone.

Re: Mpls performance on CCR1036-8G-2S.

Posted: Wed Aug 05, 2020 10:55 am
by ste
Hi Leandro,

There are some _major_ issues with MPLS in RouterOS v6. The major ones being the lack of Fast Re-Route, RSVP paths not failing over(or back), NLRI not being updated for L3VPN's and stale labels causing traffic to disappear.

It is usable for basic stuff like VPLS, but it is all done 100% in slow path (no fast path), you would be better off in this situation to just use EoIP tunnels on a OSPF routed L3 backbone.
We run MPLS/VPLS/LDP and consider to drop. Reasons:
- Stale labels. Sometimes you have a route in OSPF but packets do not get forwarded. This is triggered by address/route changes
- Debugging routing is a pain esp. with stale labels. You cant trust routing table and traceroute does not help as it does not work well with MPLS.
- Rerouting is lame
- MT does not work on this, so we might run into problems with an update and get no solution

So if EOIP is fast enough I would go this direction.

Anyone compared EOIP/VPLS Performance ?

Re: Mpls performance on CCR1036-8G-2S.

Posted: Wed Aug 05, 2020 5:19 pm
by IPANetEngineer
Hi Leandro,

There are some _major_ issues with MPLS in RouterOS v6. The major ones being the lack of Fast Re-Route, RSVP paths not failing over(or back), NLRI not being updated for L3VPN's and stale labels causing traffic to disappear.

It is usable for basic stuff like VPLS, but it is all done 100% in slow path (no fast path), you would be better off in this situation to just use EoIP tunnels on a OSPF routed L3 backbone.

I don't disagree with the comments here but that said,

I find VPLS to perform far better than EoIP on MikroTIk if the design goal is point to multipoint. EoIP was intended to be PtP and not PtMP.

As long as you understand the limitations for your use case, there is nothing wrong with running a MIkroTik based MPLS network. The platform you mentioned is one of the most common. If you're at 50% now, i'd expect the same or slightly lower.

Re: Mpls performance on CCR1036-8G-2S.

Posted: Thu Aug 06, 2020 3:16 am
by mducharme
In our case, we run VPLS instead of EoIP wherever possible. Our network does not really have redundant paths - it is like a giant tree structure, with a trunk backbone and sites that spread out as branches from there, not connected to each other. In our case, the lack of fast reroute does not impact us, because we really don't have redundant paths to fail over to in the first place.

One thing with my own personal experience with EoIP - performance can be heavily impacted by latency unless the IP MTU is allowed to be around 1542 bytes to carry a 1500 byte IP packet without fragmentation. Increasing the IP MTU on a router interface is not only going to increase the size of EoIP packets but also any other IP packets generated by the same interface. IMO, one of the strengths of MPLS is the fact that it has an independent MTU control, so that you can allow a higher MTU for tunnels without your management traffic suddenly creating packets greater than 1500 bytes.

If you do not set an IP MTU of at least 1542 on the interface that the EoIP tunnel is being formed over, the EoIP tunnel will be fragmented and the impact of any loss and latency will be amplified and reduce the TCP performance enormously.

And, RouterOS v7 with its rewritten routing engine should also significantly improve MPLS.

Re: Mpls performance on CCR1036-8G-2S.

Posted: Fri Aug 07, 2020 3:47 pm
by leostereo
Guys , thanks for your advices on this.
After reading your advices , have following conclutions:

a
Implementing mpls should not increase resourses above pure ip implementation.
b
MPLS on mokrotik with redundant links can have some problems when path changes.
MPLS on mikrotik does not support fast re route, so.
For a single triangle topology or a 5 fullmesh routers ...
How long can take network to converge on new path ? 5 secs ?
Is it possible to never recovery from path change?
c
EOIP , can be a nice option if you need to create a ptp link between locations.
Just need to be carefull with mtu.
VPLS is better for ptmp links.

Re: Mpls performance on CCR1036-8G-2S.

Posted: Sat Aug 08, 2020 4:30 pm
by IPANetEngineer
With BFD on for OSPF, you should be able to fail over in a few seconds even without fast reroute.