Re: [c-nsp] DWDM optics on 6500s
Thanks. I assume that even though the 6509-V-E is available, until the 80gig line cards and Sup are available, you'd be stuck at 40gig/slot? Chuck -Original Message- From: cisco-nsp-boun...@puck.nether.net [mailto:cisco-nsp-boun...@puck.nether.net] On Behalf Of Tony Varriale Sent: Monday, October 19, 2009 5:07 PM To: cisco-nsp@puck.nether.net Subject: Re: [c-nsp] DWDM optics on 6500s It will shortly but it won't do you any good with the existing family of sups. The 2T will be the first (and last?) sup that can push the bandwidth to all those slots. You can also reference the 6509-V-E...it's ready for 80gbps/slot. You can order that today. Note that it's a NEBS chassis. tv - Original Message - From: Church, Charles cchur...@harris.com To: Kevin Graham kgra...@industrial-marshmallow.com Cc: cisco-nsp@puck.nether.net Sent: Monday, October 19, 2009 1:12 PM Subject: Re: [c-nsp] DWDM optics on 6500s Are you saying a 6513-E chassis exists? I can't find any reference to it. That would solve a few of the problems we currently have (density issue) Chuck -Original Message- From: cisco-nsp-boun...@puck.nether.net [mailto:cisco-nsp-boun...@puck.nether.net] On Behalf Of Kevin Graham Sent: Monday, October 19, 2009 11:45 AM To: Nick Hilliard; mti...@globaltransit.net Cc: cisco-nsp@puck.nether.net Subject: Re: [c-nsp] DWDM optics on 6500s As a side issue, there are electrical limitations imposed by the physical cross-bar unit inside the actual chassis, but I don't know how much of a problem these limitations are in practice. 6500E was the key for this. Besides nutty amounts of POE capacity, it also picked up improved backplane for 20g+ fabric and extending to all 11 LC slots in the 6513. (Still need to dig up details, as faster SSO time is also tied to chassis, though I can't recall why). ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/ ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/ ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/ ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
I assume that even though the 6509-V-E is available, until the 80gig line cards and Sup are available, you'd be stuck at 40gig/slot? Correct (nothing special about the 09-V-E in this respect compared to any other the -E's as far as I know). This is the same as how the traditional (pre-E) 6500 chassis was capable of doing to do 2x20gb per slot, but it was a step ahead of the rest of the system which would only deliver 8gb (w/ SFM/SFM2) until the Sup720 was released. ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
Yup! No 80g cards yet. I haven't been debriefed on the architecture but I'm assuming it is going to be 2 x 40g or 4 x 20g backplane lanes. tv - Original Message - From: Church, Charles cchur...@harris.com To: Tony Varriale tvarri...@comcast.net; cisco-nsp@puck.nether.net Sent: Tuesday, October 20, 2009 9:14 AM Subject: RE: [c-nsp] DWDM optics on 6500s Thanks. I assume that even though the 6509-V-E is available, until the 80gig line cards and Sup are available, you'd be stuck at 40gig/slot? Chuck -Original Message- From: cisco-nsp-boun...@puck.nether.net [mailto:cisco-nsp-boun...@puck.nether.net] On Behalf Of Tony Varriale Sent: Monday, October 19, 2009 5:07 PM To: cisco-nsp@puck.nether.net Subject: Re: [c-nsp] DWDM optics on 6500s It will shortly but it won't do you any good with the existing family of sups. The 2T will be the first (and last?) sup that can push the bandwidth to all those slots. You can also reference the 6509-V-E...it's ready for 80gbps/slot. You can order that today. Note that it's a NEBS chassis. tv - Original Message - From: Church, Charles cchur...@harris.com To: Kevin Graham kgra...@industrial-marshmallow.com Cc: cisco-nsp@puck.nether.net Sent: Monday, October 19, 2009 1:12 PM Subject: Re: [c-nsp] DWDM optics on 6500s Are you saying a 6513-E chassis exists? I can't find any reference to it. That would solve a few of the problems we currently have (density issue) Chuck -Original Message- From: cisco-nsp-boun...@puck.nether.net [mailto:cisco-nsp-boun...@puck.nether.net] On Behalf Of Kevin Graham Sent: Monday, October 19, 2009 11:45 AM To: Nick Hilliard; mti...@globaltransit.net Cc: cisco-nsp@puck.nether.net Subject: Re: [c-nsp] DWDM optics on 6500s As a side issue, there are electrical limitations imposed by the physical cross-bar unit inside the actual chassis, but I don't know how much of a problem these limitations are in practice. 6500E was the key for this. Besides nutty amounts of POE capacity, it also picked up improved backplane for 20g+ fabric and extending to all 11 LC slots in the 6513. (Still need to dig up details, as faster SSO time is also tied to chassis, though I can't recall why). ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/ ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/ ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/ ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
On 05/10/2009 22:41, Mark Tinka wrote: That said, it's also clear the 6500 isn't done yet, and it's still got a number of tricks up its sleeve. The question is, Will you wait?. The c6500 is just a chassis. So, if you're referring to the trick of upgrading both the line cards and the supervisor engine to something better, then yes, it's got more tricks up its sleeve. As a side issue, there are electrical limitations imposed by the physical cross-bar unit inside the actual chassis, but I don't know how much of a problem these limitations are in practice. Perhaps the problem of getting reliable 20G+ parallel data transfers across the backplane is greater than dealing with the bandwidth limitations imposed by the electrical characteristics of the physical crossbar. Being hardware-related, this sort of stuff is well beyond my sphere of knowledge, and for all I know, c65k's operate using hoards of maxwell's daemons being slave-driven by microscopic evil pixies. Nick ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
On Monday 19 October 2009 07:42:10 pm Nick Hilliard wrote: The c6500 is just a chassis. So, if you're referring to the trick of upgrading both the line cards and the supervisor engine to something better, then yes, it's got more tricks up its sleeve. Of course, that's what I meant :-). I'd think it's implied that a chassis without any useful line cards is just a rock taking up space :-). My reference was more in terms of the platform than just a series of chassis'. Cheers, Mark. signature.asc Description: This is a digitally signed message part. ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
As a side issue, there are electrical limitations imposed by the physical cross-bar unit inside the actual chassis, but I don't know how much of a problem these limitations are in practice. 6500E was the key for this. Besides nutty amounts of POE capacity, it also picked up improved backplane for 20g+ fabric and extending to all 11 LC slots in the 6513. (Still need to dig up details, as faster SSO time is also tied to chassis, though I can't recall why). ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
Are you saying a 6513-E chassis exists? I can't find any reference to it. That would solve a few of the problems we currently have (density issue) Chuck -Original Message- From: cisco-nsp-boun...@puck.nether.net [mailto:cisco-nsp-boun...@puck.nether.net] On Behalf Of Kevin Graham Sent: Monday, October 19, 2009 11:45 AM To: Nick Hilliard; mti...@globaltransit.net Cc: cisco-nsp@puck.nether.net Subject: Re: [c-nsp] DWDM optics on 6500s As a side issue, there are electrical limitations imposed by the physical cross-bar unit inside the actual chassis, but I don't know how much of a problem these limitations are in practice. 6500E was the key for this. Besides nutty amounts of POE capacity, it also picked up improved backplane for 20g+ fabric and extending to all 11 LC slots in the 6513. (Still need to dig up details, as faster SSO time is also tied to chassis, though I can't recall why). ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/ ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
On Mon, 2009-10-19 at 14:12 -0400, Church, Charles wrote: Are you saying a 6513-E chassis exists? I can't find any reference to it. That would solve a few of the problems we currently have (density issue) We've been told about it at a local tech update about the Catalyst platform about a month ago, but I also can't find any material about it. -- Peter ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
Are you saying a 6513-E chassis exists? I can't find any reference to it. Apparently not yet. (I had never paid attention to availability, as any places we might use it would depend on full fabric connectivity). Quick search turned up (the rather depressing): http://www.cisco.com/web/AP/partners/assets/docs/Day1_03a_Catalyst_Update.pdf ...it would appear the intention is to release the chassis with the new supervisor. (Obviously the timelines cited there are out of date, since that also cites an EARL8-based 720 in '09 and VTOR, which I'm guessing we'll never see on a 6k). Several of the used vendors have matches for WS-C6513-E, so it may well be on the global price list. That would solve a few of the problems we currently have (density issue) My understanding to date is that it won't do any good until the next-gen sup is out (as presumably there would be no other reason to hold back on an 11x2 fabric configuration). ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
If you want to cut delay for switching, you may want to consider the new top-of-rack 10G boxes, which are typically cut-through. You may find I'm thinking about that for within the datacenter. It's hard finding a justification for the C-vendor's products though - a N7 is just too much, and I guess the N5 is OK but even starting out it's not... inexpensive, esp when it doesn't even include a meaningful layer-3 capability. I guess I understand the product reasoning - a large datacenter is built around N7Ks, with N5K distro - which is great, if you have a massive datacenter... but what about us poor saps in the middle and lower tiers? Or aren't we interesting anymore? Oh, I understand, we're supposed to be virtualizing and buying the 1000v switches...except virtual servers don't do me crap for good... Personally, I have a bit of a thing against X2, but that's just me. Make your own mind up. Fair enough. 3) Does 6500 switching performance blow super-hard, or just so-so hard? (6-15us is ok.) Yes a 4900M might be faster, or a J-product, but I don't want to change platform really, I need NAT and don't want to use routers, I want to keep box count down (co-lo), and having a whole box just for passing 10G doesn't IMO make sense because I'd still have to get it into the 6500 anyway. The 6500 is a great 1G switch platform, but doesn't excel in the 10G range, particularly with 6704 blades. Admittedly, for the cost, I can buy an arista 1U for wave passthru and just tap multiple 1Gs over to the 6500. Why particularly with 6704 blades? Is there something particularly wrong with them? Thanks, -bacon ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
On 05/10/2009 15:35, Jeff Bacon wrote: Admittedly, for the cost, I can buy an arista 1U for wave passthru and just tap multiple 1Gs over to the 6500. Aristas use SFP+. Good luck running colours over them. :-) Actually, Optoway in Taiwan produce CWDM SFP+ transceivers. I don't know anyone using them, but given the power constraints imposed by the SFP+ form factor, I wouldn't expect long reach or anything. Why particularly with 6704 blades? Is there something particularly wrong with them? Depends on what you do with them. They are a first generation blade, and are 6yo technology at this stage and, well, things have moved on since 2003. XENPAK is moribund as a transceiver type which means that any money you invest into buying transceivers will probably be written off when you retire the blade. If you're concerned about storm control (which personally, I am), the 6704 can only limit to 0.33% of port capacity, which means that if you get a broadcast / multicast storm on a 6704 port, it will bang out 33 megs of data before storm control even notices. Most hosts will happily ignore the multicast traffic, but the broadcast traffic could cause serious trouble. If you need to push wire-speed 10G on a 6704, there are conflicting reports as to whether this works well. Some people say yes; others no - there's lots of discussion about this in the c-nsp archives. It can help to use a DFC if you're banging out a lot of traffic, but that's extra €€€ on top of a product which already has a high cost per port. The 6708 is lots better than the 6704 if you operate it in non- oversubscribe mode, apart from anything else, it has a built-in DFC, which means that you don't need to retrofit this for high traffic environments. As I said, it depends on what you want to do. If you're running just a couple of gigs and don't care about the broadcast traffic problem or, say, are using them for L3 traffic instead of L2, then they are great. Similarly, the C65k+sup720 platform makes a really nice high density, feature rich 1G platform. But if you're planning to run lots of very high bandwidth stuff, it might be better to use a different platform. Nick ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
In order to use SFP+ from other vendors in Arista, you need to get them enabled first. -Azher Nick Hilliard wrote: On 05/10/2009 15:35, Jeff Bacon wrote: Admittedly, for the cost, I can buy an arista 1U for wave passthru and just tap multiple 1Gs over to the 6500. Aristas use SFP+. Good luck running colours over them. :-) Actually, Optoway in Taiwan produce CWDM SFP+ transceivers. I don't know anyone using them, but given the power constraints imposed by the SFP+ form factor, I wouldn't expect long reach or anything. Why particularly with 6704 blades? Is there something particularly wrong with them? Depends on what you do with them. They are a first generation blade, and are 6yo technology at this stage and, well, things have moved on since 2003. XENPAK is moribund as a transceiver type which means that any money you invest into buying transceivers will probably be written off when you retire the blade. If you're concerned about storm control (which personally, I am), the 6704 can only limit to 0.33% of port capacity, which means that if you get a broadcast / multicast storm on a 6704 port, it will bang out 33 megs of data before storm control even notices. Most hosts will happily ignore the multicast traffic, but the broadcast traffic could cause serious trouble. If you need to push wire-speed 10G on a 6704, there are conflicting reports as to whether this works well. Some people say yes; others no - there's lots of discussion about this in the c-nsp archives. It can help to use a DFC if you're banging out a lot of traffic, but that's extra €€€ on top of a product which already has a high cost per port. The 6708 is lots better than the 6704 if you operate it in non- oversubscribe mode, apart from anything else, it has a built-in DFC, which means that you don't need to retrofit this for high traffic environments. As I said, it depends on what you want to do. If you're running just a couple of gigs and don't care about the broadcast traffic problem or, say, are using them for L3 traffic instead of L2, then they are great. Similarly, the C65k+sup720 platform makes a really nice high density, feature rich 1G platform. But if you're planning to run lots of very high bandwidth stuff, it might be better to use a different platform. Nick ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/ ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
On Mon, Oct 05, 2009 at 04:06:31PM +0100, Nick Hilliard wrote: Depends on what you do with them. They are a first generation blade, and are 6yo technology at this stage and, well, things have moved on since 2003. XENPAK is moribund as a transceiver type which means that any money you invest into buying transceivers will probably be written off when you retire the blade. Don't forget they are absurdly under-buffered (16MB per card, compared to 256MB for 6708), and you can easily cause head of line blocking with certain traffic profiles. If you want to run anywhere close to line rate on them you need to monitor for drops or overruns and be prepared to play the port shuffle game to find an arrangement that works. Passing a lot of traffic within the same fabric channel (from port 1-2, or 3-4) is the biggest sin, it will start dropping at 7 Gbps. -- Richard A Steenbergen r...@e-gerbil.net http://www.e-gerbil.net/ras GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC) ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
Don't forget they are absurdly under-buffered (16MB per card, compared to 256MB for 6708), and you can easily cause head of line blocking with certain traffic profiles. If you want to run anywhere close to line rate on them you need to monitor for drops or overruns and be prepared to play the port shuffle game to find an arrangement that works. Passing a lot of traffic within the same fabric channel (from port 1-2, or 3-4) is the biggest sin, it will start dropping at 7 Gbps. Well that's wonderfully comforting. Though I really probably only need two ports anyway - ring-in and ring-out. Maybe not so bad. I'd consider a 720-VS-10G head if I had some confidence that those two ports on the sup were actually connected to the fabric. I don't really need to run line rate - this is more about latency and burst capacity than sustained throughput. I have loads that burst from 0 to 500Mb/sec (then back) in nothing flat, and multiple of those may run through the wire at the same time. Or not. Someone pointed out that the X2 and SFP+ xcvrs don't have much punch, and I'm going to be shooting 20-30km through passive MUXes. So that might matter. (This is a bit of a roll-yer-own local metro NYC ring, which I'm doing because I can get the wave for not much more than I'd pay for the switched gig.) ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
On Mon, Oct 05, 2009 at 03:47:05PM -0500, Jeff Bacon wrote: Well that's wonderfully comforting. Though I really probably only need two ports anyway - ring-in and ring-out. Maybe not so bad. I'd consider a 720-VS-10G head if I had some confidence that those two ports on the sup were actually connected to the fabric. Can't tell you anything about the VS-10G, but if you're doing it on 6704 make sure you use 1 and 3, or 2 and 4, not 1-2 etc. Unfortunately I have to deal with many hundreds of 10GE ports on 6704s (what can I say, they're cheap :P), so we tend to try to pair them up as port-channels (i.e. members 1/1 and 1/2, 2/1 and 2/2, etc) since this guarantees traffic will never go in port 1 and out port 2 on any given fabric channel. I don't really need to run line rate - this is more about latency and burst capacity than sustained throughput. I have loads that burst from 0 to 500Mb/sec (then back) in nothing flat, and multiple of those may run through the wire at the same time. Or not. Yeah ok that won't challenge pretty much any hardware. :) Someone pointed out that the X2 and SFP+ xcvrs don't have much punch, and I'm going to be shooting 20-30km through passive MUXes. So that might matter. X2 is nothing more than a physically smaller XENPAK case, the interface and for the most part the components (if you take apart a modern XENPAK, you'll see most of it is empty space) are exactly the same. Basically X2 only exists so lazy companies who don't want to redesign their boards (Hi Cisco!) can keep using the same components from their old XENPAK designs. SFP+ is an entirely different beast, two generations removed from XENPAK (XENPAK-XFP-SFP), and with very low max power caps which prevent it from being used for most long reach/DWDM applications. Basically SFP+ only exists so you can stuff 48 10GE ports into a blade or 1U switch, but it's really only useful if you need to do a large number of short reach ports (i.e. datacenter aggregation). The only redeeming quality of SFP+ is you can finally get LR for them (I won't touch SR outside of same-rack applications, way too many problems) at not unreasonable prices. XFP is still the best all-around optics platform for the full range of features, but unfortunately you'll see less and less focus here as everyone jumps on the SFP+ bandwagon as the next new thing even when it is completely unnecessary and infact only serves to limit function. Slightly dated now (from feb 08) but mostly still accurate: http://www.nanog.org/meetings/nanog42/presentations/pluggables.pdf -- Richard A Steenbergen r...@e-gerbil.net http://www.e-gerbil.net/ras GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC) ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
Well that's wonderfully comforting. Though I really probably only need two ports anyway - ring-in and ring-out. Maybe not so bad. I'd consider a 720-VS-10G head if I had some confidence that those two ports on the sup were actually connected to the fabric. The 10Gig ports on the VS-S720 are fabric attached. I don't really need to run line rate - this is more about latency and burst capacity than sustained throughput. I have loads that burst from 0 to 500Mb/sec (then back) in nothing flat, and multiple of those may run through the wire at the same time. Or not. I think lots of people are in the latency not bandwidth situation. That's probably why most vendors aren't producing dense 10Gig cards yet. We have the situation where GigE latency is too high for some apps, but 10Gig is okay. Someone pointed out that the X2 and SFP+ xcvrs don't have much punch, and I'm going to be shooting 20-30km through passive MUXes. So that might matter Opnext claims ER SFP+. Haven't seen anyone doing ZR or anything more exotic yet. Sure it will come though. Something FEC/EFEC in the 200km range would be interesting for many people. -- Tim: Sent from New York, NY, United States ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
We've selected the 6708 for our 10Gig installs. DFCs and good sized buffers. Lots of availability on the used market. Can be run in line-rate or over-subscribed mode, which might suit your deployment. I have hopes for SFP+ linecards to drive 10Gig costs down, but I don't think much is going to happen until 40Gig/100Gig is the new backbone. Tim: On Mon, Oct 5, 2009 at 4:47 PM, Jeff Bacon ba...@walleyesoftware.com wrote: Don't forget they are absurdly under-buffered (16MB per card, compared to 256MB for 6708), and you can easily cause head of line blocking with certain traffic profiles. If you want to run anywhere close to line rate on them you need to monitor for drops or overruns and be prepared to play the port shuffle game to find an arrangement that works. Passing a lot of traffic within the same fabric channel (from port 1-2, or 3-4) is the biggest sin, it will start dropping at 7 Gbps. Well that's wonderfully comforting. Though I really probably only need two ports anyway - ring-in and ring-out. Maybe not so bad. I'd consider a 720-VS-10G head if I had some confidence that those two ports on the sup were actually connected to the fabric. I don't really need to run line rate - this is more about latency and burst capacity than sustained throughput. I have loads that burst from 0 to 500Mb/sec (then back) in nothing flat, and multiple of those may run through the wire at the same time. Or not. Someone pointed out that the X2 and SFP+ xcvrs don't have much punch, and I'm going to be shooting 20-30km through passive MUXes. So that might matter. (This is a bit of a roll-yer-own local metro NYC ring, which I'm doing because I can get the wave for not much more than I'd pay for the switched gig.) ___ cisco-nsp mailing list cisco-...@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/ -- Tim: Sent from New York, NY, United States ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
On Monday 05 October 2009 11:06:31 pm Nick Hilliard wrote: As I said, it depends on what you want to do. If you're running just a couple of gigs and don't care about the broadcast traffic problem or, say, are using them for L3 traffic instead of L2, then they are great. Similarly, the C65k+sup720 platform makes a really nice high density, feature rich 1G platform. But if you're planning to run lots of very high bandwidth stuff, it might be better to use a different platform. From Cisco, I think that if the goal is to aggregate n x 10Gbps Ethernet in abundance, for pure Layer 2 core switching over a limited distance (within the data centre), the Nexus 5000 might not be such a bad consideration. Preliminary pricing for this vs. a couple of WS-X6708 is comparable, and better in certain cases. YMMV. That said, it's also clear the 6500 isn't done yet, and it's still got a number of tricks up its sleeve. The question is, Will you wait?. Cheers, Mark. signature.asc Description: This is a digitally signed message part. ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
[c-nsp] DWDM optics on 6500s
I am looking at getting some metro waves (mostly 20-40km) between sites; I'm working with a provider who is using passive splitters on dark runs and they're willing to split me out a wave for near the same cost as just running a gig switched. Currently, I am thinking of using a 6704 blade with a DFC3B (I'm using sup7203Bs - no call for the C model in my environ) and buying tuned optics. The goal is primarily serialization latency delay reduction, not actually running 10G of traffic - I'll be lucky to run 1-2GB (though it'll mostly be 60-100byte packets). 1) The cisco optics appear to be in short supply and damn expensive. I may have to go third-party. I know it's a gamble. Any other issues I Should think about besides what's been discussed here? 2) Is XENPAK on 6704 viable? Any gotchas I should know about with XENPAKs vs X2? 3) Does 6500 switching performance blow super-hard, or just so-so hard? (6-15us is ok.) Yes a 4900M might be faster, or a J-product, but I don't want to change platform really, I need NAT and don't want to use routers, I want to keep box count down (co-lo), and having a whole box just for passing 10G doesn't IMO make sense because I'd still have to get it into the 6500 anyway. 4) What's reliability on the tuned optics (vs say using a freq-shift box with a normal 10G optic)? Is this of the level that I should expect significant BER, or keep a spare on the shelf, or is it pretty much rock-solid? As you might guess, this is my first foray into actually implementing DWDM runs - I've studied it and planned it, but this is now at the buy stuff level, and I don't want to assume I know everything. Relevant document pointers appreciated. Thanks, -bacon ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] DWDM optics on 6500s
On 02/10/2009 17:44, Jeff Bacon wrote: I am looking at getting some metro waves (mostly 20-40km) between sites; I'm working with a provider who is using passive splitters on dark runs and they're willing to split me out a wave for near the same cost as just running a gig switched. I went through this some while back, and on the basis that: - coloured xenpaks are exotic, expensive and only produced by a single manufacturer in the world (opnext, as you ask) - coloured xenpaks will only last as long as your 6704 card, meaning that when you retire this kit, your entire coloured optics investment is lost - transponders were not hugely more expensive than the prices I was quoted for cisco coloured optics ... I decided that coloured xenpaks, while marginally cheaper in the short term, were actually a bad strategic move in the long term. Given the way that our network has changed since we made that decision, it turns out that it was a good decision to make, as we're completely flexible about what kit we use at each end of the link, and have chosen to exercise that flexibility. There is also a much better selection of coloured XFPs on the market than coloured xenpak. The goal is primarily serialization latency delay reduction, not actually running 10G of traffic - I'll be lucky to run 1-2GB (though it'll mostly be 60-100byte packets). If you want to cut delay for switching, you may want to consider the new top-of-rack 10G boxes, which are typically cut-through. You may find that these boxes + SR SFP+ + wdm transponders is quite cost favourable compared to c6500 chassis space + 6704 + coloured xenpak. Cisco N5K may be a good option here. But other vendors have similar style boxes (Brocade Ti24X, Extreme X650, F10 S2410, Arista Networks *.*, etc). Oh, and the SFP+ boxes will also run 1G ethernet on SFPs (although the N2K has some limitations). This is a nice feature win. 1) The cisco optics appear to be in short supply and damn expensive. I may have to go third-party. I know it's a gamble. Any other issues I Should think about besides what's been discussed here? coloured xenpaks are a single vendor product and I have heard that they are mostly made to order, hence the delay. 2) Is XENPAK on 6704 viable? Any gotchas I should know about with XENPAKs vs X2? X2 == xenpak version 2. Their power draw is slightly less than xenpak, but lots more than xfp / sfp+. Only HP and Cisco use X2 for ethernet switches - everyone else uses XFP and latterly SFP+, which means that there is less pricing pressure and and more vendor lock-in if you go down the X2 route. Personally, I have a bit of a thing against X2, but that's just me. Make your own mind up. 3) Does 6500 switching performance blow super-hard, or just so-so hard? (6-15us is ok.) Yes a 4900M might be faster, or a J-product, but I don't want to change platform really, I need NAT and don't want to use routers, I want to keep box count down (co-lo), and having a whole box just for passing 10G doesn't IMO make sense because I'd still have to get it into the 6500 anyway. The 6500 is a great 1G switch platform, but doesn't excel in the 10G range, particularly with 6704 blades. Nick ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/