Configuring BGP on WS-3750G-12S-E

[Praful Soni]

Dear Friends,

I have WS-3750G-12S-E with me and would like to configure BGP on the same. I want your suggestion / help whether it is advisable to configure BGP on 3750G or not.

Please suggest if anybody has already deployed BGP on 3750.

Thanks in advance for your support

[Leo Laohoo · Accepted solution]

BGP????  On a 3750G-12S?????

I don't think so.

You'll need a router to do that.  This is because 3750G switches don't have the memory space to store the BGP table, even partial.

[Praful Soni]

Thanks Leolaohoo !!!

Anybody else can also share experience please.

[paolo bevilacqua · Accepted solution]

Leo is, of course, correct.

Thank you for the nice rating and good luck!      

[Leo Laohoo]

Thanks, Paolo.

[Reuben Farrelly · Accepted solution]

Saying that you'll need a router to run BGP isn't accurate or true.  The 3750s (all flavours) can run BGP just fine.

3560s/3750s have enough memory to run several hundred BGP prefixes or more, if that meets your needs.  That may be sufficient to run an IGP, or to advertise BGP to an upstream provider (and just take a default router via BGP in the downstream direction).  Definitely won't take a full Internet table, but in many cases that's not necessary anyway.

Been there done that a few times.  It works fine, and as always you just need to be mindful of the limitations of the platform in so far as memory and TCAM size.  You will need the IPSERVICES featureset though, as BGP isn't supported on the IPBASE featureset.

Featurewise it's fairly much all OK, I can't remember anything missing in the implementation compared to much bigger switches and the config is exactly the same as all other IOS devices.

Edit:  Found this:

http://www.cisco.com/en/US/products/hw/switches/ps5023/products_tech_note09186a00801e7bb9.shtml

which indicates your 3750G-12S should be able to take up to 20,000 routes IN HARDWARE provided you use the right SDM template (and trust me, you do not want to exceed this, ever).  You will still need to make sure you don't eat up DRAM memory with too many routes.   I would suggest staying well below that limit.

Message was edited by: Reuben Farrelly - with more information

[Praful Soni]

Hi Farrely,

I really appreciate your right guidance. I gone throght the documentation on site and got same feedback from other expert also.

I have already done BGP configuration in my 3750 with limited advertisement of downstream routes.

Thanks again.

[paolo bevilacqua]

While the post above, as well the documents linked are technically correct, in saying one can BGP on a 3750, even will the due cautions, that does not good advice and is not a best practice.

The reason is very simple, entry level layer 3 switches have limited memory and a very slow CPU. So the periodic BGP table scans, any path re-calculation, and the general overhead, will make that the switch will be likely be burdened by high CPU and memory usage, with all the possible negative consequence, that in the worst case can lead to service disruption.

So you one can ask, it that sholn't be done, why Cisco has this feature in the code? The answer is that as everything else in life, you are given the tool, then what use you make of it, it's up to you.

The real solution is to design the network correctly, and place a true router where it is has to be placed, e.g. at the BGP peering point.

So, the advice by Leo was correct, and that makes the difference between professionals speaking from experience and full understanding of things, and others from theoretical and web browsing "knowledge".

[Reuben Farrelly]

My comments are based on first hand experience both of my own doing this over the last 10+ years, and of seeing others deploy these switches in relatively large deployments.  Not just "web browsing knowledge".

Provided an engineer for a particular deployment, appropriately sizes and scales these switches (ie does not expect them to take thousands of routes, or large routing tables of any protocol) I still maintain there is no issue either fundamental or practical in deploying limited scale BGP on them.

I know of and have worked for some medium sized ISPs here in Australia where this is exactly how they have been deployed.  [Personally I'd much rather run ME series switches in those environments, but that's somewhat beside the point].

[Reuben Farrelly]

Use case as an example of my point - a customer who I worked with early last year had two Gigabit Internet connections to two separate ISP's.  They needed failover from one to the other (so it was a simple case of using only one Internet connection at once).  They had their own /24 which we used conditional BGP advertising to select which ISP to advertise to.  They only required a default route in from both upstreams and only advertised one prefix out to the upstreams.

Now to do that on a router we'd need at least an ASR1k minimum in order to handle Gigabit throughput.  The cost for that is at 10x more than a 3560/3750.  It only needed to handle two prefixes - and no NAT, no ACLs.   Are you suggesting that in this case the customer should have put a router in, even though a switch like this was perfectly adequate for the job?

[paolo bevilacqua]

Reuben,

You made your case clear, thank you. Different engineers can tackle problems different ways. Personally I don't see a reason to continue discussing this.