GSR, 7600, Juniper M?, oh my!
Hello all, I'm faced with a difficult decision. I work for a large multi-node regional ISP (and Cisco shop). In our largest nodes we've found the Cisco 7500 series routers to be at the end of their useful life due to the throughput generated by POS OC-3 feeds and 10,000+ broadband users whose traffic needs to be moved out of the node. Short of building a farm of 7500's the need to upgrade seems clear. But where to go? The Cisco GSR platform seems a logical choice, but their new 7600 series units are attractive for their cost. Juniper may also have a place at this end of the processing spectrum. I'd also like to ensure that the new platform supports doing CAR and ACLs at line rate, given the client base. I wanted to see what other operators in this situation have done, so I would appreciate anyone's input or insight into the pros and cons of these platforms or any other ideas as to how I can grow beyond the Cisco 7500.
I'm faced with a difficult decision. I work for a large multi-node regional ISP (and Cisco shop). In our largest nodes we've found the Cisco 7500 series routers to be at the end of their useful life due to the throughput generated by POS OC-3 feeds and 10,000+ broadband users whose traffic needs to be moved out of the node. Short of building a farm of 7500's the need to upgrade seems clear.
Will the interfaces likely continue to be POS OC-3 ? What is the growing path for this: POS OC-12, GigE ?
But where to go? The Cisco GSR platform seems a logical choice, but their new 7600 series units are attractive for their cost. Juniper may
also
have a place at this end of the processing spectrum. I'd also like to ensure that the new platform supports doing CAR and ACLs at line rate, given the client base.
The GSR line-cards to what you want would need to be the "edge" ones, based on either Engine 3 or Engine 4+. 7600 requires WAN cards to support POS, I think GSR and Juniper M are more likely candidates for this design. Rubens
On Tue, Jan 06, 2004 at 01:11:25PM -0500, bcm wrote:
Hello all,
I'm faced with a difficult decision. I work for a large multi-node regional ISP (and Cisco shop). In our largest nodes we've found the Cisco 7500 series routers to be at the end of their useful life due to the throughput generated by POS OC-3 feeds and 10,000+ broadband users whose traffic needs to be moved out of the node. Short of building a farm of 7500's the need to upgrade seems clear.
But where to go? The Cisco GSR platform seems a logical choice, but their new 7600 series units are attractive for their cost. Juniper may also have a place at this end of the processing spectrum. I'd also like to ensure that the new platform supports doing CAR and ACLs at line rate, given the client base.
I wanted to see what other operators in this situation have done, so I would appreciate anyone's input or insight into the pros and cons of these platforms or any other ideas as to how I can grow beyond the Cisco 7500.
7500s? In 2004? Throw those things in the trash where they belong. It's always amazing to me how many people will cling to obsolete things for years just because it is what they know. Even a Juniper M5 will do 16 OC3's with line rate filtering and forwarding. There are probably a dozen design considerations based on requirements you haven't described, but if you're doing primarily sonet, 7600 isn't really the way to go. -- Richard A Steenbergen <ras@e-gerbil.net> http://www.e-gerbil.net/ras GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)
7500s? In 2004? Throw those things in the trash where they belong. It's always amazing to me how many people will cling to obsolete things for years just because it is what they know.
Even a Juniper M5 will do 16 OC3's with line rate filtering and forwarding. There are probably a dozen design considerations based on requirements you haven't described, but if you're doing primarily sonet, 7600 isn't really the way to go.
I usually agree with RAS, but not this time. 7500's have a place; not everyone is looking for wire speed at 2499384 gigglebits. In our network, 7500's have made a home for the leaf-ends of DS3's in dial and DSL pops, handling 20 to 40 megs/sec with easy (ie, vip2-50's at 15% CPU). Also handle MPLS AToM with ease, PPPOE/A if needed (we choose to segregate that onto VXR's). 7500's also can handle many ChDS3's with ease. And, also, with RSP16/VIP8, lots of traffic can be handled. All at a substantially less price than even the cheapest used M5 you can find. 7507 + dual ps + rsp4 can be had for $1000 to $1500, and VIP2-50's can be had for $300. And then, you can use all the PA's you have laying around from your 7200's. Secondly, 6509 + OSM is actually a sweet solution, and provides way more LAN aggregation than any traditional 'router' can. Wire speed, too. (disclaimer: this is coming from someone who has all m5, m10, m20, and m40 core, with 6509's, 7500's and 7200's strewn all over his network. These are my opinions, and probably differ from people who are indentical to me).
On Tue, Jan 06, 2004 at 08:25:18PM -0500, Alex Rubenstein wrote:
7500s? In 2004? Throw those things in the trash where they belong. It's always amazing to me how many people will cling to obsolete things for years just because it is what they know.
Even a Juniper M5 will do 16 OC3's with line rate filtering and forwarding. There are probably a dozen design considerations based on requirements you haven't described, but if you're doing primarily sonet, 7600 isn't really the way to go.
I usually agree with RAS, but not this time.
7500's have a place; not everyone is looking for wire speed at 2499384 gigglebits. In our network, 7500's have made a home for the leaf-ends of DS3's in dial and DSL pops, handling 20 to 40 megs/sec with easy (ie, vip2-50's at 15% CPU). Also handle MPLS AToM with ease, PPPOE/A if needed (we choose to segregate that onto VXR's).
Yes, I suppose 7500s do have a place. To quote rs, "I have a lace doily and a lamp atop a 7500; it's quite nice." :) Your argument applies if you have them sitting around already, don't need performance, don't need filtering, don't need stability, AND have enough places to put lamps or plants at your house, or places to sit at the colo. That's fine and all, we all have legacy gear we have to do something with, but the thought of someone considering building new pops of 7500s still sends chills down my spine. Maybe I'm just biased because I actually do need all of the things mentioned above, and I already have a very nice empty M160 chassis for a chair at the colo, thus making the existance of a 7500 anywhere near me more of a burden in trash disposal fees than anything else. If you really do have a situation where you don't need or want any of the aforementioned qualities, you can use Linksys for all I care. :)
7500's also can handle many ChDS3's with ease. And, also, with RSP16/VIP8, lots of traffic can be handled.
All at a substantially less price than even the cheapest used M5 you can find. 7507 + dual ps + rsp4 can be had for $1000 to $1500, and VIP2-50's can be had for $300. And then, you can use all the PA's you have laying around from your 7200's.
The cost of the extra rack space will be more than you paid for the entire 7500 every few months. If I'm going to run many ChDS3, I'd rather go ahead and future-proof myself (I could end up replacing those PICs with OC12 QPP's doing ds1's some day) for a little bit more money now, still be able to run 20+ ChDS3 ports with plenty of uplink capacity in a 3U M10, and save the money in the cost of the space and power. -- Richard A Steenbergen <ras@e-gerbil.net> http://www.e-gerbil.net/ras GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)
On Tue, Jan 06, 2004 at 08:25:18PM -0500, Alex Rubenstein wrote:
find. 7507 + dual ps + rsp4 can be had for $1000 to $1500, and VIP2-50's can be had for $300. And then, you can use all the PA's you have laying around from your 7200's.
Oh, also, on the subject of used market pricing... It's been a while since I looked at Cisco ChDS3 PA pricing in any serious detail, but as I recall they were valued as though they were made of gold and personally blessed by Pope John Chambers when compared to used Juniper ChDS3s. If this is really your application, you could probably sell your load of ChDS3 PAs to the waiting crowd of suckers on eBay and trade up to a Juniper with money left over, on any decent number of chds3's. -- Richard A Steenbergen <ras@e-gerbil.net> http://www.e-gerbil.net/ras GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)
At 09:37 PM 1/6/2004, you wrote:
Oh, also, on the subject of used market pricing...
It's been a while since I looked at Cisco ChDS3 PA pricing in any serious detail, but as I recall they were valued as though they were made of gold and personally blessed by Pope John Chambers when compared to used Juniper ChDS3s. If this is really your
Last time I bought a PA-MC-T3, I paid $4700.00 and that is a GREAT price. They usually sell for $5k+ Two years ago, I sold some extras for $1500-2000/ea. They have more than doubled in price. The PA-MC-2T3 cards are going for $10-12k used now! I bought them in 2000 and 2001 for $4500 each.
application, you could probably sell your load of ChDS3 PAs to the waiting crowd of suckers on eBay and trade up to a Juniper with money left over, on any decent number of chds3's.
So... Everyone always says Juniper is so great. How does one get a legitimate copy of JunOS and a software support contract. I have called and emailed Juniper at least 3-4 times via each method and never received any response regarding getting a license for a used router. I would like to buy a used M5/10/20/40 just to play with it so I can learn more about them and how they work. We are interesting in the scalability and the cheap PIC cards available for the Juniper gear - especially the channelized DS3 interfaces. Does anyone have a useful contact or number for software maintenance at Juniper? I don't want to spend $5-10k on used Juniper gear if I can't get an OS to run on it. Any Juniper lovers care to help? TIA! -Robert Tellurian Networks - The Ultimate Internet Connection http://www.tellurian.com | 888-TELLURIAN | 973-300-9211 "Good will, like a good name, is got by many actions, and lost by one." - Francis Jeffrey
7500s? In 2004? Throw those things in the trash where they belong. It's always amazing to me how many people will cling to obsolete things for years just because it is what they know.
Don't agree with this. For E1/T1 access these boxes are fine. Yes they are long in the tooth but they are quite capable. I wouldn't spend a huge amount of time or money trying to make them do anything else though.
Even a Juniper M5 will do 16 OC3's with line rate filtering and forwarding. There are probably a dozen design considerations based on requirements you haven't described, but if you're doing primarily sonet, 7600 isn't really the way to go.
Depends on what "primarily sonet" means. Regards, Neil.
On Tue, 6 Jan 2004, bcm wrote:
But where to go? The Cisco GSR platform seems a logical choice, but their new 7600 series units are attractive for their cost.
7600's have all the craptacularity of 6500 switches, because thats what they are, I would reccomend against them. Buying GSR's is probably the right replacement for 7500's if you want to stick with Cisco.
Juniper may also have a place at this end of the processing spectrum. I'd also like to ensure that the new platform supports doing CAR and ACLs at line rate, given the client base.
Then you'll be wanting J boxes then, cuz Cisco doesn't do that very well from my experience.
GSRs are useless if you are doing any kind of aggregation. Their traffic shaping abilities are embarrassing. 7500 is the classic aggregator. They do the job quite well, actually. Based on cost right now, I would take 10 7500s over 1 7600 anyday. For transit, though, I would use a Juniper - hands down. Cisco has to get their $hit together in terms of software stability.... one would hope it would get more stable over time.... but alas no. Dan. "Tom (UnitedLayer)" wrote:
On Tue, 6 Jan 2004, bcm wrote:
But where to go? The Cisco GSR platform seems a logical choice, but their new 7600 series units are attractive for their cost.
7600's have all the craptacularity of 6500 switches, because thats what they are, I would reccomend against them.
Buying GSR's is probably the right replacement for 7500's if you want to stick with Cisco.
Juniper may also have a place at this end of the processing spectrum. I'd also like to ensure that the new platform supports doing CAR and ACLs at line rate, given the client base.
Then you'll be wanting J boxes then, cuz Cisco doesn't do that very well from my experience.
GSRs are useless if you are doing any kind of aggregation. Their traffic shaping abilities are embarrassing.
Historically yes, but no longer. The latest line of GSR cards now give them much greater capability in this area even though it was never designed as an access box.
7500 is the classic aggregator. They do the job quite well, actually. Based on cost right now, I would take 10 7500s over 1 7600 anyday.
If you are just aggregating E1/T1 then I'd agree, but the minute you need DS-3/E3/STM-1/ATM/100BaseT/Gige aggregation then the 7600 is a far better choice cost wise, if only the code on it worked. The feature set for the 7600 and the roadmap is very impressive but Cisco need to spend a large part of time a: fixing the bugs in the code and b: training their people how this box works.
For transit, though, I would use a Juniper - hands down. Cisco has to get their $hit together in terms of software stability.... one would hope it would get more stable over time.... but alas no.
"transit" ? In my view the software on the GSR is very stable, in fact probably one of the most stable parts of the IOS family. I can count on one hand the number of real bugs we have for the GSR in the last 3-4 years, and two of them were security related. [which as I understand it Juniper is not ammune to either]. There are issues with the early engine 0 and engine 1 cards on the GSR so beware if buying second user cards. Also I would not underestimate the effort of going multivendor for core and access devices. There are alot of issues other than just cost when going multivendor. Regards, Neil.
Back to the original question.. A lot of your decision comes down to what you're going to be doing with the box and when you expect your next jump from OC3 to OC12(or greater). Also, you need to consider your comfort level with JUNOS vs IOS. If you're cool with JUNOS then multiple M series boxes are worth investigating. Our experience with them has been almost nothing but positive, plus they will allow you to expand to greater than OC3, providing you with some future proofing. 7600's have proven to be fine boxes, especially if you have need for Ethernet port density at the same layer as your optical circuits. A lot of feature support is going to depend on your supervisor/msfc selection. If you go this route, and the coffers are full, check out the new(er) sup720's. However, based on your ACL and Policing requirements, the Sup2/MSFC2 combo should be sufficient. Also, keeping in mind the emergence of point to point Ethernet solutions in the WAN/MAN (ie Metro Ethernet, and MPLS and L2TPv3 pseudowires) keeping Ethernet at your edge might prove useful one day. The GSR, IMHO, is a higher tier box based on both it's scalability to OC192 and cost. Since you're just going to OC3's now, I doubt the GSR will be your best bet for the cost, but then again I haven't priced one out lately. If you're really pinching pennies, then check out upgrading your 7500's with RSP8/16s and faster VIPs. But, if you're putting multiple OC3's on a box, then your down links will likely start turning to GE. I'd stay away from the GEIPs if possible. And for your 7200's, look into the NPG-G1 which have line rate GE ports onboard. We've used them and they are pretty solid. A head to head GRE bakeoff between the NPE-G1 and an RSP8(with dCEF) proved the NPE to be far superior. Josh -----Original Message----- From: owner-nanog@merit.edu [mailto:owner-nanog@merit.edu] On Behalf Of bcm Sent: Tuesday, January 06, 2004 1:11 PM To: nanog@merit.edu Subject: GSR, 7600, Juniper M?, oh my! Hello all, I'm faced with a difficult decision. I work for a large multi-node regional ISP (and Cisco shop). In our largest nodes we've found the Cisco 7500 series routers to be at the end of their useful life due to the throughput generated by POS OC-3 feeds and 10,000+ broadband users whose traffic needs to be moved out of the node. Short of building a farm of 7500's the need to upgrade seems clear. But where to go? The Cisco GSR platform seems a logical choice, but their new 7600 series units are attractive for their cost. Juniper may also have a place at this end of the processing spectrum. I'd also like to ensure that the new platform supports doing CAR and ACLs at line rate, given the client base. I wanted to see what other operators in this situation have done, so I would appreciate anyone's input or insight into the pros and cons of these platforms or any other ideas as to how I can grow beyond the Cisco 7500.
If you choose the 7600's I would highly recommend going with the Sup720's the price difference is not that great and they incorporate the SFM which gives you the option of running dCEF on your WAN cards. Scott C. McGrath On Thu, 8 Jan 2004, Josh Fleishman wrote:
Back to the original question..
A lot of your decision comes down to what you're going to be doing with the box and when you expect your next jump from OC3 to OC12(or greater). Also, you need to consider your comfort level with JUNOS vs IOS. If you're cool with JUNOS then multiple M series boxes are worth investigating. Our experience with them has been almost nothing but positive, plus they will allow you to expand to greater than OC3, providing you with some future proofing.
7600's have proven to be fine boxes, especially if you have need for Ethernet port density at the same layer as your optical circuits. A lot of feature support is going to depend on your supervisor/msfc selection. If you go this route, and the coffers are full, check out the new(er) sup720's. However, based on your ACL and Policing requirements, the Sup2/MSFC2 combo should be sufficient. Also, keeping in mind the emergence of point to point Ethernet solutions in the WAN/MAN (ie Metro Ethernet, and MPLS and L2TPv3 pseudowires) keeping Ethernet at your edge might prove useful one day.
The GSR, IMHO, is a higher tier box based on both it's scalability to OC192 and cost. Since you're just going to OC3's now, I doubt the GSR will be your best bet for the cost, but then again I haven't priced one out lately.
If you're really pinching pennies, then check out upgrading your 7500's with RSP8/16s and faster VIPs. But, if you're putting multiple OC3's on a box, then your down links will likely start turning to GE. I'd stay away from the GEIPs if possible. And for your 7200's, look into the NPG-G1 which have line rate GE ports onboard. We've used them and they are pretty solid. A head to head GRE bakeoff between the NPE-G1 and an RSP8(with dCEF) proved the NPE to be far superior.
Josh
-----Original Message----- From: owner-nanog@merit.edu [mailto:owner-nanog@merit.edu] On Behalf Of bcm Sent: Tuesday, January 06, 2004 1:11 PM To: nanog@merit.edu Subject: GSR, 7600, Juniper M?, oh my!
Hello all,
I'm faced with a difficult decision. I work for a large multi-node regional ISP (and Cisco shop). In our largest nodes we've found the Cisco 7500 series routers to be at the end of their useful life due to the throughput generated by POS OC-3 feeds and 10,000+ broadband users whose traffic needs to be moved out of the node. Short of building a farm of 7500's the need to upgrade seems clear.
But where to go? The Cisco GSR platform seems a logical choice, but their new 7600 series units are attractive for their cost. Juniper may also have a place at this end of the processing spectrum. I'd also like to ensure that the new platform supports doing CAR and ACLs at line rate, given the client base.
I wanted to see what other operators in this situation have done, so I would appreciate anyone's input or insight into the pros and cons of these platforms or any other ideas as to how I can grow beyond the Cisco 7500.
A lot of your decision comes down to what you're going to be doing with the box and when you expect your next jump from OC3 to OC12(or greater). Also, you need to consider your comfort level with JUNOS vs IOS. If you're cool with JUNOS then multiple M series boxes are worth investigating. Our experience with them has been almost nothing but positive, plus they will allow you to expand to greater than OC3, providing you with some future proofing.
One additional data point is that due to excess forwarding resources its possible in JUNOS 6.x to condense many physical 760x and 750x down in to 1 or 2 M series using ithe virtual router capability. In a nutshell each virtual router is its own administrative domain so you can have different staff maintain different virtual routers on the same physical box. Fewer physical boxes might be an advantage for some. See release notes for JUNOS 6.x in the tech pubs area of the juniper.net web site for more details on virtual router features. -Rob
participants (12)
-
Alex Rubenstein -
bcm -
Dan Armstrong -
Josh Fleishman -
neil＠DOMINO.ORG -
Randy Bush -
Richard A Steenbergen -
Rob Healey -
Robert Boyle -
Rubens Kuhl Jr. -
Scott McGrath -
Tom (UnitedLayer)