Cisco Support Community
Showing results for 
Search instead for 
Did you mean: 

Welcome to Cisco Support Community. We would love to have your feedback.

For an introduction to the new site, click here. If you'd prefer to explore, try our test area to get started. And see here for current known issues.

New Member

BGP + 2 ISPs on 2620...?

I need to set up a web cluster which has connectivity to two ISPs (redundancy as well as performance reasons). There will be somewhat significant traffic present, but not a significant budget. I only have access to 2 standard 2620s each with a T1 WIC.

-- Should I put the 2 WICs in one 2620 and use BGP?

-- Can a 2600 handle this kind of setup (I hear BGP tables are 100MB+)? If so, what does it need, more memory, different WIC?

-- Can I set up one for each ISP and then have the 2620s talk to each other, to share the load so to speak? The cluster will be behind a PIX, by the way, don't know if that helps this kinda setup.

-- Any sample config snippets would be greatly appreciated, I'm new to BGP...

VIP Purple

Re: BGP + 2 ISPs on 2620...?

If you want redundancy, then use both 2620's. I would suggest you consider running HSRP between two of the Ethernets that connect to your network betwen the 2620's.

If you have one AS number, then I would talk BGP between each 2620, and its respective ISP, and use something like internal BGP between the two routers.

New Member

Re: BGP + 2 ISPs on 2620...?

As already suggested - HSRP should be used between the two 2620's on the LAN facing the PIX. Likewise the PIX points its default route to the HSRP address. The two 2600's will each have a BGP session to its wan connected ISP, and another to the other 2620 (two peers on each router). You must have registered IP address space (typically provider independent) and an AS number before the ISP's will play ball. As you have assumed - The technical issues will be memory related if you really want to load share outbound traffic. One possible approach may be to reduce memory overheads by limiting the prefixes learned inbound by using prefix lists and setting default route to cover the stuff you filter out. In theory for example you could start by only accepting /16 or shorter masks, and see if the resulting tables fit in your routers memory. Make the prefix length shorter and shorter as needed. While you’re ignoring more routes at least you'll be able to make reasonable route selection to the major networks, following default for everything else.

The major pitfall for BGP for beginners is that they often don't know that must add static routes (or learn dynamic routes via the internal routing protocol) for the WAN links that connect the ISP's so the both routers know how to reach the next-hop for all learned BGP routes. Without this BGP will not accept routes learned from the other 2620 because the next-hop will not be known. An alternative is to use next-hop self on both ends of the internal bgp session. Example BGP configs can be generated free on by putting in the peering relationship type (ISP being transit provide), AS numbers, and ip address of the peers. This tool however does not show an example of the prefix-list idea that may be used reduce the number prefixes you accept. You'll have to look in CCO for examples of this. This step will only be required if have limited memory. Since you will point a default route to each ISP you can trim down the routes you accept to fit within your available memory and gain some benefit from running BGP for selecting your outbound path. Alternatively you could just announce your network to both isp's and accept nothing from them, ONLY following default route for all outbound traffic. Your HSRP primary should then be set to the better performing ISP. For announcing your network, simply put a static pointing to your firewall and a network statement that matches the mask size under the BGP process. The examples from above include standard filters to block rfc1918 addresses. By comparing the examples, and reading up about the syntax on CCO you should be able to learn a lot about BGP to give you an idea of the options. If you have time to test in a lab, you can test the BGP examples between your two routers and create static routes to represent the BGP routes, again the tools on the above site can help create the config for this. CCO is of course your best reference for understanding how to configure BGP. You could also obtain "Internet Routing Architectures" from Amazon if you want to get into the topic in depth. I hope this helps. Good luck.

New Member

Re: BGP + 2 ISPs on 2620...?

The 2620s can be used if you are using static routes to point out to the Internet.

But since you have conectivity to two ISPs, you will most likely need to run BGP.

In this case the 2620s are not recommended. You will need a minimun of a 3640 routers especially if you will be receiving the full Internet routes. There are currently almost 18,000 network addresses in the Routing table. You need an ABSOLUTE MINIMUN of 64Megs but at least 128Megs to run sufficiently. Running BGP with two ISPs will provide you a higher level of redundency and ease to configure load sharing traffic between the ISPs.


New Member

Re: BGP + 2 ISPs on 2620...?

If you don't care about getting all of the Internet's routes then you can use BGP to advertise your routes to the internet and then use static routes with the same metric to route traffic outbound. The static routes will have to point to your serial interfaces. As long as you have route-cache turned off, you will do a round-robin distribution of your outbound traffic. Otherwise you are going to need a bigger router with a bunch more memory and then run EBGP and IBGP like the other suggests suggested :)

Hope everything works out :)