ASR9000 4GB vs 8GB Memory

[ctychan]

Anyone know the difference between A9K-RSP-4G and A9K-RSP-8G?  What is the usage for additional 4GB memory?

[Sergei Vasilenko]

Hello,

Just the memory size diff for a higher control plane / service scalability (routing tables / VPN tables,

lots of interfaces, complex policies, etc).

http://www.cisco.com/en/US/prod/collateral/routers/ps9853/data_sheet_c78-500699.html

Sergey

[Ivan Krimmel]

the main bottleneck is the line cards though.

[ctychan]

If so, any figure for routing table size (and others) for RSP with 4GB and 8GB memory?

If ASR9K is based on N7K technology, N7K only have max 1M routing table size with XL license.

Sergey Vasilenko wrote:
Hello,
Just the memory size diff for a higher control plane / service scalability (routing tables / VPN tables, 
lots of interfaces, complex policies, etc).
http://www.cisco.com/en/US/prod/collateral/routers/ps9853/data_sheet_c78-500699.html
Sergey

[Sergei Vasilenko]

Hello,

For route scale the l3xl scale profile general definition could be found here:

http://www.cisco.com/en/US/docs/routers/asr9000/software/asr9k_r4.1/system_management/configuration/guide/sysman_cg41asr9k_chapter1.html#task_3A082F6CD31D4A238070C3CD7279E67A

...snip...

In general, the route switch processor (RSP) with 4  GB of memory is capable of loading 1.3 million IPv4 routes. However,  consider using an 8 GB RSP if the router needs to hold a large number of  routes.

...end snip...

Regards,

Sergey

[xthuijs · Accepted solution]

Allow me to add some explanations here.

Some of the chips used in N7K are reused in a different fashion on the a9k. While the fabric operation is similar between N7K and A9K the hardware forwarding is substantially different.

In terms of routing, there are 2 things to keep separate the RIB (routinginformation base) and FIB (forwarding information base).

the RIB holds all routes from all protocols and the FIB is the derived set on which we are actually forwarding. Routes in here have won the AD contest and best metric etc.

IT is the FIB that gets downloaded to the LC's forwarding structures in the hardware processors. That size is limited to 512k/1M/1.3M depending on which scale profile you choose.

this number is a hard limit of the NP (network processor) and is increased with the next gen hw aka Typhoon coming out end of this year to 4M routes.

Now, you can imagine that if you have a route reflector, or a large L3 VPN network etc, that thenumber of PATHS is veyr large, it may not mean that all paths are actually inserted in the FIB. But you just want to hold them in the RIB.

that is where the 4G/8G discussion comes into play. If you have a high MSE scale or require multidimentsional scale in terms of many pmaps, large RIB, lots of BGP routes etc etc then the 8G version is something to go for.

If you are looking at the 4G and only for BGP routes, depending on the paths, attributes etc, you *can* scale up to 7M routes there in BGP already. Again, this depends on the paths and its attributes, if 2 paths have the same set of attributes they consume less memory then when they have different atts.

Hope that clarifies things a bit.

Xander Thuijs - CCIE #6775

Sr Tech Lead ASR9000

[Leo Laohoo]

is increased with the next gen hw aka Typhoon coming out end of this year to 4M routes.

You mean in the next two weeks?

[xthuijs]

That is 100% correct! the NEW typhoon based linecards with the next generation NPU's have a capability of 4M routes! a near 400% capacity upgrade of todays FIB tables in the Trident NPU's.

thanks

xander

Xander Thuijs - CCIE #6775

Sr Tech Lead ASR9000

[Leo Laohoo]

Looking forward to it!  Thanks for the heads-up.