Hi,

I'm looking for someone who has experience with the Cisco 2948G-L3.

We were looking at building out a new network presence. We will be starting with a GigE fiber from verio, and blending in Level 3 within the next year as it grows.

Initially, I was requesting quotes for 6500 stuff, and comming back at $10K to $15K each. With hot spares, thats almost double to have a reliable network. Definatly right in line for what you get, but quite a bit overkill for only moving a couple hundred meg/sec.

In the course of my adventures, I was pointed to the 2948G-L3. Its a layer 3 switch that supports BGP protical. The idea would be to get the network up and get traffic established. Once it passes a couple hundred meg, I can justify the 6500. BUT, if I can run a reliable network off of the 2948G-L3 up to a reasonable amount of traffic I think it would be a good investment to start smaller and build up larger later.

Any thoughts, suggestions, experience with this unit?

OOOHH YAA..
a product link
2948G-L3 product page (http://www.cisco.com/en/US/products/hw/switches/ps606/products_data_sheet09186a008009267f.html)

If you're going to be running a GigE, and BGP... you should go with the 6500 series.

Umm..

Read his post next time. He is looking for an alternative to paying $15k for a loaded 6500. Doh!

I will probably get slain for suggesting this, but I have been reading up on Summit 48i, I know its capable of the task, you can get it with 256mb ram (correct me if I am wrong), etc.

Only thing I dont know is... how much bandwidth can you push through it?

Other suggestions are a Cisco 7206 w/ 256mb ram.

Just wait, someone will come along and tell you that you should let switches switch and routers route, but uhhh... whats a Cisco 6500 series??? A SWITCH THAT ROUTES. How about a 12000?

The line between a router and a switch grows thinner every day. I am doing some research myself, and I admit to not having done this before as well, so I dont recommend anything that I just said :)

hmmkay :)

There was a large discussion on this topic earlier.

My feelings is the summit it probably a bit under the task. My hope the cisco would be a better running start, to lay the groundwork for a redundant BGP routed fiber network.

Verio has Junipers and 6509's in the buildings, I would most likely just be another port off there 6509, feet away.

I'm not pullin in OC trunks, but I need to lay the ground work for routing, including blocks from arin, etc.

But back to the topic, what is the limits of the 2948G-L3? Will I get a decent run?

2948G-L3 ~$2600
6506 ~$15,000

For the cost of one 6506, I could buy 2 2948's AND ~20 intermediate 1U servers.

Perhaps check with ImageStream(http://www.imagestream-is.com/) and see if they have a router that fits your routing needs.(And, desired budget.)

Then you can get a Layer 2 switch.

Perhaps I've looked at routers and switches for months, and this looks to be the most profitable way to run a clean network? Companys that over build are where all the cheap used cisco gear comes from.

Just a question JTY, have you ever pulled $15,000 out of your pocket, cash? When you do, think you might look at a few options?

I guess the odds of me getting a reply from someone who has acually used it and knows and from the arm chair techs who read all the threads are about 80 to 1

I certainly understand, your desire to look at various options. You may consider asking Verio for a recommendation. They may be willing to point you in a direction.

Also, you could try emailing Cisco. They'd be able to tell if it will hold up to the task.

And, no I've not pulled 15k out of my pocket. But, I do look at options no matter what I'm spending.

Here's a review on the 2948G-L3, http://www.nwfusion.com/reviews/2000/0529rev.html

Verio has offered to include a router with the connection, but that doesn't mean it's going to be best for me.

I'm sure Cisco would say that equipment would perform as advertised. With the IOS, rate limiting, vlan's it should give a decent run. Remember, I'm building a presence, not moving established traffic immediatly.

BGP was the icing on the cake. And I would understand if it wouldn't route a large amount of traffic. But if it can route a decent amount of traffic thats money in the bank.

Hmm, there are a few network engineers around WHT. Perhaps one of them will read this thread.

How much total bandwidth are you planning on pushing through it? I know you mentioned a couple hunderd meg, earlier.

The 2948G-L3 would be something I would buy to do switching anyway, but to a point it might be good enough to route a small amount of traffic too. I don't know its limits when you try to BGP route it and add a ACL list, etc.

If its 10M, then I don't need another layer 3 switch right now. If its 100M, then we have something interesting.

That was an interesting link. The fact they test and compair it as a closet switch doesn't inspire confedence ;)

I've been trying to find some performance numbers for you. But, I'm not finding much.

How much total bandwidth are you planning on pushing through it? I know you mentioned a couple hunderd meg, earlier.

I'm fishing to see the limits, and if theres another piece of equipment in that range that can deliver a good flow.

I don't have a limit set in my mind, but I think you see where I'm taking this now. If I save a few months time of having a 6506 sitting in a rack moving 20M, and at the same time can still string in the lines, setup DNS, file servers, backup devices, basically move in. The real catch is the routing and BGP.

Thanks JTY for helping. Honestly, at first I thought you were trolling ;)

I try to help. :) Just trying to get a feel for what you wanted from the hardware.

I imagine, that pushing 30-40MB with BGP it'll do just fine. Just keep an eye on it, and watch it's CPU load.

The more I look into it, the less I like it. Looks like a awesome layer3 switch though.

I'm hearing the same things from alot of people, not enough intermediate gear for this application :(

I keep ending up at a 6506 . 30Mbps would be just enough to get me in trouble, and the very last thing I want is to overload the router and provide crappy service. My clients pay for and expect quality, I wouldn't get by if this thing overloads intermittently.

do *not* do this, its suicide. i will hold off on lecturing you on how you should use routers for routing and switches for switching. besides that, there are a few points that require cluing in.

with switching, the performance limits are the switching fabric and the packets per second limit. although i see them advertised at 22gbps and 10mpps, i seriously doubt the pps rate figures.

with layer 3, the performance limits come (roughly/basically) from the amount of ram required to store routing data and the way the device handles flows (full asic, some sw, all sw). while the switch may be good enough to do a few hundred meg just doing switching, it will most likely max out well before then doing layer 3. not sure if you will be able to get it to hold full routes from two providers, not to mention that routing with a decent l2 switch that they enhanced to do l3 so they could put it on the feature list (and not much else) may not be the greatest idea. read: it cant handle flows efficiently and will die a horrible death.

look at the product page. cisco describes it as an aggregation switch. it will do fine having your distrib switches plugged into it, but l3 is out of the question imho.

its inexpensive, so get one to do your l2 stuff. if you need more, get something bigger later. dont worry about bgp for now, since you dont have a second link yet. when you can afford it, you will be able to afford a decent box to handle that as well (if your business plan works of course =).

good luck,
paul

Acually, I have a second link, but its a private agreement. Level 3 would be a third, but I'm not willing to anwer question about it so I didn't mention it.

Unfortunatly, I don't have any agreements in place with level 3 currently, so to aggrate bandwidth here will take a nice initial agreement that will probably be best doing after establishing a good income on the network. Quality is just so expensive ;) I'm aready dishing out a buttload of cash, on top of whatever gear is put in place.

So since using this gear is such a bad idea that was pumped buy sales 'engineers' , what would you recommend putting in for a router?

My business plans work, thats why this is all being bought with cash, and not credit ;) But I do expect 3-6 months of bleeding out of my ear before I can flip it to a positive cash flow. Good thing I have a diverse company with income from multiple sources ;)

3550 if your going to be cheap, 7200 vxr if you're not...

ICQ me and we can take up our old tradition of a coney breakfast at 5am and some chit chat 8-)

Basically, I've found that its worth just going 6500 or not doing it at all. Price vs performance, it seem to be the ticket right now.

I just picked up a nice modular routing switch, I'll pick up the BGP in a routing solution later. Last thing I want to do is build out too much too fast.

We can meet up for a technical discussion, IIRC it's your turn to buy breakfast.

Get that baby!:)

Get that baby!:)

Can I point out that although it says it supports BGP it's not specific about what version? You'll be needing BGP4 if you have any plans of multihoming your network. You'll also need an AS number which will cost you.

Routing switches are dodgy, it also depends on which mode they use for switching, a 7204 would be a nice router, a vxr even nicer, Extreme's stuff is also nice, but please try and use a router for routing.

Winkie/rusko,

I know rsanders situation a little better then you, and in this scenario, a switch would be of more value then a router, because of the number of vlan's, traffic, and ip situation...

BTW, a 6509 with a MSFC is essentially a switch with a router on the backplane, much better then using a external router...

If the funds are available for a 6509, it is about the best option you have...

Being a engineer myself I see many things that need to be considered.

Make sure you have a design that will

1. Accomodate short term and long term goals.
2. Allow for equipment failure
3. Scale
4. Be modular

Most people make the mistake of trying to build networks in a non-modular non-layered fashion. This creates performance issues, reliability issues and increases capital expenditure in the long run.
This being said a specific model or manufacturer is not your solution, a good design is.
If you used a 6500 for everything what happens when your primary circuits start flapping and eat up CPU and free mem on the 6500 causing routing protocols to time out and customers to lose connectivity? What about maintenance? Everything tied into one device creates a single point of failure. Not to mention the number of anomalies that could result.
It's usually best pratice to keep your edge rouers seperate from your core/distribution switches.

Other things I know for a fact:
The MSFC on a 6500 CANNOT handle BGP. You can go ahead and enable it but look out. Cisco recommends using A SUP 2 with a MSFC2 as a minimum.
Quote from Cisco:"We recommend a minimum of 128 MB of RAM in the router to store a complete global BGP routing table from one BGP peer. On the Catalyst 6000 receiving a full BGP table, it is recommended to have MSFC2 with 256 MB of RAM to avoid bug CSCdt13244. The memory consumption by BGP routes depends on the number of attributes such as multi-path support, soft reconfiguration, number of peers, and AS_PATH. For more details on the BGP memory requirement, refer to RFC 1774 .
"
http://www.cisco.com/en/US/tech/tk365/tk80/technologies_tech_note09186a0080094a83.shtml

MSFC2's cost 10+k.

My advice would be to get a dedicated router for your initial circuit and route down to a layer 2 switch.
A Cisco 7200 series with a NPE300 should run you 3k used.
It is designed to handle BGP, HSRP, OSPF and all other routing functions. Then hang a good layer 2 switch off it for your customers to connect to. This would be a simple and cost effective design. In the future when you get your second provider you can buy a second 7200. At this point you can setup HSRP on both your routers so all customers hanging off the layer 2 switch will have redundancy between each edge router. You can then trunk the two edge routers to distribute traffic evenly and for additional failover between provider A and provider B.


7200's with NPE300's and full BGP tables can handle about 200 megs throughput. Don't let anyone tell you different. Since you'll have 2 in the future this does not mean you can do 200 on each. This means you can do 100 on each because if one fails the other needs to handle everything.

Then if you need to upgrade in the future you can swap out the 7200's with 7500's. 7500's are cheap and versitile... and the layer 2 switch can be moved down the design model at this point and a 5500 or 6500 can be put between the 7x00 and the layer 2 access switch.

Originally posted by Papa Smurff
Being a engineer myself I see many things that need to be considered.

Make sure you have a design that will

1. Accomodate short term and long term goals.
2. Allow for equipment failure
3. Scale
4. Be modular


These are extremly important.. There is always something to say for doing things right the first time. I know right now you will not likely see the need, but having to completely redesign your network later on will be even worse when you have large amounts of traffic going through it and bringing it completely down isn't an option...

No one person here can tell you exactly what you need. I would suggest bringing in a network engineer to look over your current system, tell them your budget, and ask for suggestions or a full out plan. Many will work per hour and give you a very nice, feasable design..

When building the back bone of your internet business its good to spend the extra money on the inital setup :D


Example: Not meaning to slam on Pweb/DedicatedNow but it was talked about a few months ago that they may be going to a VLANed network. Which would require a complete IP renumber for the servers going from standard to VLAN.. If this would have been thought of initally it wouldn't have been an issue and there wouldn't have been a 10 page thread here with customers about ready to die :D

Originally posted by The Prohacker
These are extremly important.. There is always something to say for doing things right the first time. I know right now you will not likely see the need, but having to completely redesign your network later on will be even worse when you have large amounts of traffic going through it and bringing it completely down isn't an option...

No one person here can tell you exactly what you need. I would suggest bringing in a network engineer to look over your current system, tell them your budget, and ask for suggestions or a full out plan. Many will work per hour and give you a very nice, feasable design..

When building the back bone of your internet business its good to spend the extra money on the inital setup :D


Example: Not meaning to slam on Pweb/DedicatedNow but it was talked about a few months ago that they may be going to a VLANed network. Which would require a complete IP renumber for the servers going from standard to VLAN.. If this would have been thought of initally it wouldn't have been an issue and there wouldn't have been a 10 page thread here with customers about ready to die :D


Amen.

I'd agree that even just a 6500 is not a option.

A 7206VXR (or better yet, a Juniper m20) with Cisco 2948s cascaded behind it wouldn't be to bad of an option.

I was involved in a purchase of a 6509 to replace a series of 2948s that we had on our network, it was a good replacement, however the company only did layer 2 stuff, the 6509 was plugged into a GSR as well.

Now I work with Juniper M20s with switches behind these also doing pure layer2, this also works very well.

But I would *definately* not recommend even thinking about trying to do BGP on a 2948, or even on a 6509. Just a bad thing to think.

^^ a 7204 should be fine, there's no need to ramp up to a 7206vxr :)

Also, as Prohacker mentioned, you need to make sure all customers you set up are assigned valid subnet blocks. You do not want to have to renumber customers or interfaces in the future. Make sure to consider this in your design and to reserve a few IP's open in each customer subnet for your use. Never know when of if you'll need them and you sure as heck don't want to redesign everything.

a 6509 for BGP would be quite fine, I have been using one in an enviroment far more complex then a data center, and it works wonderfully, and yes, you do need MSFC2, and they can be had on ebay in a 6509, populated for $20k

A 6509 with MSFC2 is essentially the same as a 6509 with a 7204 as well... router on the backplane...