Web Hosting Talk : Web Hosting Main Forums : Colocation and Data Centers : good switch for BGP

[hpham]

I don't have any experience in router management. So, I want to have a switch can handle this
- the total bandwidth is less than 200mbps
- BGP sessions from 2 upstreams: my current network from datacenter, and a new one (I am thinking to get Level(3) or Internap or Verizon)

Any recommendation please?
Thanks.

[bsolaris]

You need a router for BGP.
So you are eventually going to have 2 200mb/s ports or 2 100mb/s ports?
Both links will be rj45 ethernet or will they be something else?

Are you currently doing BGP now? Do you have an AS #?

[stephenM]

You can only run BGP on a very small amount of switches, and even then all you'll be able to store is default routes - you won't be able to store a full global routing table on them which is what BGP is all about. It would be advisable to buy a proper BGP-capable router.

[WII-Aaron]

There are lots of switches that will do what you need. Foundry XMR is a good choice. If you're looking to go cheap you can get a used Extreme BD 6808 with an MSM-3 on Ebay. That will hold 2 full tables or if you want to spend more the Cisco 6500 or 7600 Catalyst switches with a SUP720-3BXL will both work.

To hold 2 full BGP tables you need a switch with at least 512MB for the tables. if you're not doing full tables then a smaller unit will do.

If you want to go into the router arena, the Cisco 12000 series or Juniper M10 would suit you fine.

[DoCk]

If you don't need full tables a Foundry FESX is a good fit.

[Spudstr]

Originally Posted by DoCk

If you don't need full tables a Foundry FESX is a good fit.

cisco SUP2U's are cheap too if you just want to do default routes, they might hold a full routing table right now but they are going to be dying out in the near future. of course with this you need to get a 6500 chassis as well..

[Zitibake]

It depends on what you want BGP for. For some people, just defaults is fine.

Some people may just need default plus 1-hop provider routes. For example, if you advertise to your customers that you have a connection to Verizon, then it's hard to explain to your customer why traffic from your net to a Verizon IP takes Level3. If you use Vz and L3, and the destination IP is on Sprint, there is usually not much difference in which route you take. I'm seeing 863 one-hop routes from Vz right now (^701$), and less than 1,000 1-hop routes from other carriers, so this kind of route table is pretty small.

One step better is to carry multi-hop routes, but to filter-out routes which pass through any Tier-1 carrier you don't connect to. If you purchase transit from Verizon and ATT, then the benefit from carrying a route to a Verio IP is modest; but if the destination is a Tier-2 connected to Vz and Sprint, then you'd be better-off knowing to hand the traffic to Vz, and not ATT. By filtering-out the "Tier-2->Sprint->ATT" route (along with other routes that pass through ATT or any other Tier-1 you don't connect to), you can probably reduce your routing table by about 80%.

You can also filter further by filtering on RIR allocations; it requires some ongoing maintenance.

There are ways to get useful functionality out of BGP, whether your limit is 7,000, 64,000 or 205,000 active prefixes.

[RoryErickson]

There are lots of options out there, I think it's really going to depend on your specific needs--and planning towards your soon to be needs.

You mention that you're going to maintain two upstreams. What sort of advertisements do you expect to be receiving? Full routes, full routes plus default, or just a default route?

Will this device also be acting as a distribution and access layer? That is, will there be customer segments hanging off it meaning you'll need ports to support them. If so, what media types will they use(ie: copper, SMF, MMF)?

What port sizes and media will your upstreams require?


--

Now, without any additional information I'd assume you're going you're likely going to using a default route only--unless you're versed in BGP the full routes really aren't very useful to you.

I'm not a big fan of the Foundry L3 gear(bad experiences with it). I'd lean more towards Cisco or Juniper.

If you're new, I'd probably suggest Cisco. A Cisco WS-C3550-12G would suit you well if you're not taking full routes and is inexpensive--they're also EOL, so don't expect new IOS releases for them. They have 10 gbic, and 2 copper 10/100/1000BaseT ports. Not a beefy box, but they can handle basic bgp and are pretty solid. If you want expandability, but have a small budget a Cisco 6500 series router will work well.

Juniper is also a great option, even something as small as a Juniper M10 or M20. They're modular, but I'm not sure they'll be able to handle the full route table for long--I don't know the memory capacities on either off the top of my head.

At any rate, if you can give more detail it would be easier to give an adequate(and useful) suggestion.


Cheers,

Rory

[FastServ]

For 200Mbps and default routes, either foundry FESX or cisco 3750G should do fine. If you plan on doing larger bandwidth down the road (e.g. >1Gbps), the FESX will far outperform the Cisco 3750 not to mention it has 10Gb ports.

[chrisbyrd]

Originally Posted by WII-Aaron

If you're looking to go cheap you can get a used Extreme BD 6808 with an MSM-3 on Ebay.

We have found (painfully) that the best position for a BD 6800 in a datacentre is either as chair or a coffee table

[hpham]

Sorry for late reply in this.

It seems something new for me here. Can anyone direct me to document of "default routes" and "full routes" please? What are the difference between them?

Thank you.

[eger]

Originally Posted by FastServ

For 200Mbps and default routes, either foundry FESX or cisco 3750G should do fine. If you plan on doing larger bandwidth down the road (e.g. >1Gbps), the FESX will far outperform the Cisco 3750 not to mention it has 10Gb ports.

I'm using a FESX and pushing over 1000mbps on it with 3 BGP peers (default routes) and it works great. Handles load much better than the comparable Cisco 36xx/37xx. It also can load balance equal cost default route BGP links with ease.

[KDAWebServices]

What model FESX, as AFAIK there are now two distinct lines with differing specs in the FESX line.

[eger]

This one is the FESX448 with the Premium upgrade. The Premium upgrade, which is a physical chip they send you to install (you get to open up your switch - yay!), enables all the advanced layer 3 stuff on it. I think the premium upgrade is for all the FESX line and I thought they all had similar specs just different port layouts.... But I'm not 100% sure.

The thing handles some crazy load though with about 40 VLANs and 8-9 /24 subnetted to VLANs traversing it.

[FastServ]

Originally Posted by eger

This one is the FESX448 with the Premium upgrade. The Premium upgrade, which is a physical chip they send you to install (you get to open up your switch - yay!), enables all the advanced layer 3 stuff on it. I think the premium upgrade is for all the FESX line and I thought they all had similar specs just different port layouts.... But I'm not 100% sure.

The thing handles some crazy load though with about 40 VLANs and 8-9 /24 subnetted to VLANs traversing it.

Yes, I was referring to the FESX448-PREM and FESX424-PREM. Both can handle silly amounts of traffic, even with multiple partial tables (100k+ routes). And down the road when your BGP needs are increasing and you upgrade your core, these devices make excellent aggregation/distribution switches with their 10GE uplinks.