guestions regarding multiple routing engines

m4rtin

I had a chance to play a little with a M20 which had two routing engines. Am I correct, that routing-engines use simple 100BASE-TX interface to communicate with each other? At least if I removed one of the RE's and executed "request routing-engine login other-routing-engine",I received following error message:

{master}
root> request routing-engine login other-routing-engine    
re0: No route to host
re0: No route to host

{master}
root>

re0 seems to indicate to Realtek 100BASE-TX NIC..

In addition, I removed routing-engine from this M20 and router was able to send packets between different networks for 45 seconds. This should be because forwarding table is in forwarding engine. However, has anyone tested this? What time did you get? And is it possible to force forwarding table to be persistent(not to check routing table for updates) for examle 10 minutes?

Aldur

First of all, you do realize that you are using a router that is probably 13 years old, right? The M20 came out in 1999 and has been EOL'd (end of life'd) for quiet a while.

I not privy to the internal physical architecture of the M20 router, but I can tell you that communication between routing engines happens via TNP over an internal address only used by the router for routing engine to routing engine communications. So, this is where your "re0: No route to host" message is coming from.

By saying that you removed the routing engine, I am assuming you removed the master routing engine. Please correct me if I am wrong. And did you not configure the router to automatically switch between routing engines on routing engine failure? If not I would advise you to have a look at the following docs for reference on how to do so.

Understanding Routing Engine Redundancy on JUNOS Routers - JUNOS 9.5 High Availability Configuration Guide

Configuring Routing Engine Redundancy - JUNOS 9.5 High Availability Configuration Guide

I have done much testing with routing engine failovers, and in my testing, when the config is correct, the transition occurred without issue and there was no packet loss. Now, if the router goes headless, both routing engines fail, the forwarding plane will continue to forward traffic until the info in the forwarding plane ages out.

You want to make the forwarding table to be persistent??? Don't you see the problems with this? There is a large potential to blackhole traffic when you have a router that is not receiving route updates but continues to forward traffic, especially if you are using any dynamic routing protocols.

And whats with the rolling eyes emoticon?

m4rtin

Aldur wrote: »

First of all, you do realize that you are using a router that is probably 13 years old, right? The M20 came out in 1999 and has been EOL'd (end of life'd) for quiet a while.

Sure. On the other hand, RE redundancy on M20 should be exactly similar to newer platforms like M10i or even MX960.

Aldur wrote: »

I not privy to the internal physical architecture of the M20 router, but I can tell you that communication between routing engines happens via TNP over an internal address only used by the router for routing engine to routing engine communications. So, this is where your "re0: No route to host" message is coming from.

What is a TNP? However, the re0 seems to be the hostname of first routing engine according to /etc/hosts.junos. Almost every PIC, FPC seems to have it's unique IP address.. Am I correct?

Aldur wrote: »

By saying that you removed the routing engine, I am assuming you removed the master routing engine. Please correct me if I am wrong. And did you not configure the router to automatically switch between routing engines on routing engine failure? If not I would advise you to have a look at the following docs for reference on how to do so.

Understanding Routing Engine Redundancy on JUNOS Routers - JUNOS 9.5 High Availability Configuration Guide

Configuring Routing Engine Redundancy - JUNOS 9.5 High Availability Configuration Guide

I have done much testing with routing engine failovers, and in my testing, when the config is correct, the transition occurred without issue and there was no packet loss. Now, if the router goes headless, both routing engines fail, the forwarding plane will continue to forward traffic until the info in the forwarding plane ages out.

Yes, I removed master RE and GRES was not configured.

Aldur wrote: »

You want to make the forwarding table to be persistent??? Don't you see the problems with this? There is a large potential to blackhole traffic when you have a router that is not receiving route updates but continues to forward traffic, especially if you are using any dynamic routing protocols.

Of course this can last only for a very little time. For example in situation where one needs to add SDRAM to RE this might be useful. One will ask the PFE to be in persistent state, pull out the RE, add memory modules, insert the RE and wait it to boot up. Should take <5min. Of course the fine solution would be to insert second routing engine at first and configure GRES. Or in case there is a router with just one RE(for example Juniper M5).. So in rare cases this might be useful, don't you think?

Aldur wrote: »

And whats with the rolling eyes emoticon?

Sorry

Aldur

m4rtin wrote: »

Sure. On the other hand, RE redundancy on M20 should be exactly similar to newer platforms like M10i or even MX960.

The M20 might have similar RE redundancy as the new platforms, but there are some major differences. For example, the MX and T series routers use a control board (CB) that controls all RE functions. Also, with an M20 you probably can only run 8.x, and there probably have been some software updates in that regards in the last 4 years. It's safe to say that they might be similar in regards to RE redundancy, but there are going to be significant differences too.

m4rtin wrote: »

What is a TNP? However, the re0 seems to be the hostname of first routing engine according to /etc/hosts.junos. Almost every PIC, FPC seems to have it's unique IP address.. Am I correct?

I copied this from the Junos High Availability book

"The protocol used for this IPC is Trivial Network Protocol (TNP). TNP is a Layer 3
protocol (like an IP) and uses Ethernet II encapsulation. Like any Layer 3 protocol, it
uses source and destination addresses, and it forms and maintains neighbor relation-
ships using Hello messages. The TNP Hello message contains Hello timers and dead
intervals, which are used to discover the failure of other hardware components (REs,
Packet Forwarding Engines or PFEs, and Flexible PIC Concentrators or FPCs). While
you cannot configure most of the TNP Hello timers, you can configure the Hello and
dead time intervals between two REs through the command-line interface (CLI)
keepalive command."

You can VTY into each FPC and PIC using TNP, so I believe that they do receive their own internal addressing.

m4rtin wrote: »

Of course this can last only for a very little time. For example in situation where one needs to add SDRAM to RE this might be useful. One will ask the PFE to be in persistent state, pull out the RE, add memory modules, insert the RE and wait it to boot up. Should take <5min. Of course the fine solution would be to insert second routing engine at first and configure GRES. Or in case there is a router with just one RE(for example Juniper M5).. So in rare cases this might be useful, don't you think?

Yea, I can see your point here. I'm not aware of anything that allows for it, but let me do some checking to see what I can find.

froggy3132000

You do that by doing GRES and NSR, you can play flip-flop with your RE's. And I have 9.5 code running on my M20.