BGP Routing for ISP Multihoming

[venkie_pai]

Hi All,

We are planning on multihoming our Internet and I would need some help on deciding the best way to set it up & also what hardware to consider.

Currently have a 100M link for our Internet with one ISP. But due to issues over the last 7 months with this provider, we have decided to have a 50M redundant link with another ISP – in an active/standby setup. Note: we might wish to upgrade the b/w across these links in the next 6-12 months to going up to 200M or 500M.

Besides normal web usage, we use the internet for a number of things – Office365, remote access VPN, site-to-site VPN’s and also to receive certain feeds. Hence, it is critical that we have our public subnet (& ASN) be available across both providers. I understand that for multihoming, we require;

1. To have full route tables from both ISP’s – which leads to significant overhead for storing route tables, & also CPU for processing etc.

OR

 2.  We have full route table from Primary ISP & default + local routes from secondary ISP, AS pre-pend and ISP to change local preference (which will serve as the backup) – which seems to be a better option.

I would like to get your thoughts on;

• Which is a better solution – option 1 or option 2?

• How much memory does the whole BGP internet routing table require? And does multihoming with 2 ISP’s require 2x that much memory? (Current stats from Hurricane Electric shows there are 687,913 IPv4 Prefixes Observed)

• Hardware to use? Would it be better to use a L3 switch like a Cisco 3850 (4G DRAM) or a router like an ASR 1000 series (ASR 1001 – 4GB/8GB).

I understand that these ASR’s are built to handle such processing and

With 4-GB memory, it can handle up to 500,000 IPv4 or 500,000 IPv6 routes

With 8-GB or 16-GB memory up to 1,000,000 IPv4 or 1,000,000 IPv6 routes etc.

But wouldn’t the 4GB memory on a 3850 switch be able to handle similar number of routes with respect to its memory? How much of the 4GB is actually allocated to TCAM in the 3850’s? And CPU?

Thanks in advance.

[e.ciollaro · Accepted solution]

Hi Venkie,

BGP full routing table is not required in your scenario: you can receive the default route from both ISP and using weight or local preference to force your router to use the primary ISP.  Advertisement to the backup ISP will be prepended to make the backup ISP worst then the primary (usually your AS will be prepended 3 to 5 times).

If you don't want to change your access routers when you'll upgrade the lines, I suggest at least a 3945E but consider that:

• router performance greatly decrease if you configure NAT, IPSec, Firewalling and so on. To have an idea look attached data-sheet
• 50 to 500 Mbps is a huge increase of bandwidth; for a 50Mbps link the 1900 family could be enough  so carefully consider the amount of bandwidth you really require and timeline for upgrade  and/or renting router's from ISP (this way when you  upgrade the lines ISP will upgrade router)

Bye,

enrico.

PS: please rate if useful

[venkie_pai]

Hi Enrico,

Thanks for responding. Okay, so just accepting default routes from each of the ISP's and using AS Path prepend to influence inbound & local preference for outbound will do the job of using one ISP as primary & the other as a backup in case the primary fails. That makes it a lot easier. Out of curiosity, if just default routes work fine even with Multihoming, why would a small to medium size enterprise want to have the entire BGP routing table? Is it so they have more control over routes to their destination etc.?

Coming back to my second question, since I am accepting only default routes, wouldnt a L3 switch do the same job? We have a few 3560CX's (WS-C3560CX-8PC-S - 512M of DRAM) which are pretty new but arent being used. Surely they can handle the current bandwidths we plan on implementing 100M & 50M? 

Regards,

Venkie

[e.ciollaro]

why would a small to medium size enterprise want to have the entire BGP routing table?

Yes, in my experience it's unusual but sometime also a little company would like to have a better control over path selection, for example using an ISP for everything but AS directly connect to the second ISP. 

Coming back to my second question, since I am accepting only default routes, wouldnt a L3 switch do the same job? We have a few 3560CX's (WS-C3560CX-8PC-S - 512M of DRAM) which are pretty new but arent being used. Surely they can handle the current bandwidths we plan on implementing 100M & 50M? 

I think that a better design is to use a router; routers have more features related to WAN and routing and this design separates device that has different goal in your network

Bye,

enrico