I'm putting together a proposal for a greenfield 10GigE network (including 10GigE uplinks to providers) while trying to be careful about budget.

One of the hardware configurations I'm evaluating that would be relatively inexpensive is a blade chassis with a 10GigE uplink module that gives me a set of 2 CX4 connectors + 1 SFP+ connector (there's no good way to substitute a fiber module in this configuration without exploding the budget).

While the SFP+ would be fine IF the datacenter supports SFP+ (which is still pretty new), CX4 obviously isn't going to work for uplinks to providers, and we'll need more than one 10GigE uplink.

HP offers CX4->Multi-Mode Fiber adapters for their ProCurve gear for less than $1000/each. This would be fine, but the problem is the CX4 adapters aren't coming out of HP/ProCurve gear. I can't seem to find non-ProCurve adapters.

So there are several inter-related questions here:

  1. Do the ProCurve adapters work with any CX4 connector?
  2. If not, are there "generic" adapters out there that do?
  3. Are there likewise any SFP+ -> MMF adapters available in case the provider doesn't support SFP+ yet?


Edit: Couple extra details:

  • The blade system is from SuperMicro, and their network modules are visible here: http://www.supermicro.com/servers/blade/networking/
  • I'm trying to avoid having to buy an external 10GigE switch to uplink the blades to.
  • We will eventually probably have to go 10GigE top to bottom (sticking Infiniband modules in the blades for use as 10GigE adapters) and switch to SuperMicro's 10GigE pass-through module with external 10GigE switches, but I want to put that off until the project is showing a return on investment.

Edit II: chris's answer set me straight on SFP+ (for unknown reasons, I'd confused it with 10GBASE-T in my mind). I'm still hoping for a pointer at CX4->MMF converters, however, since one or two 10GigE uplinks through SFP+ won't be enough long-term -- I'm looking to eventually push close to 30gbps out per chassis.


I'm not sure I understand your question.

I just looked at the manual for the supermicro blade chassis and it supports two network modules, so you'd be able to have 4 cx4 and 2 sfp+ interfaces in your chassis; this ought to provide a reasonable level of redundancy.

If your provider is offering you 10gigE over multimode fiber, you could connect the supermicro switches to your provider directly using an SFP+ module such as this one.

SFP+ isn't an interconnect standard. It is a transceiver standard to allow a single device to optionally support any of the available interconnect standards in use (long haul or short haul optical, short distance digital interconnects, etc). 10GBASE-CX4‎, 10GBaseT, 10GBASE-SR, and so on -- these are interconnect standards and each end of a link needs to use the same standard, but it doesn't matter if one end is xenpak and the other is sfp+ or hardwired, or made out of cheese, so long as the media supports the proper interconnect standard.

So I'm not sure why you need anything else like a media converter.

  • No idea how, but I seem to have gotten SFP+ and 10GBASE-T confused in my mind (very odd since none of the sources I've been looking at seem to conflate them in any way). Thanks for setting me straight there. :) Unfortunately, it doesn't entirely solve my problem -- redundancy matters, of course, but what I'm mostly concerned with right now is bandwidth, and while I'll only need one 10GigE uplink at first, over the next year I expect to need at least 3 or 4 per chassis. 2 SFP+ with fiber transceivers won't do it, so I'm still hoping someone can point me at CX4->MMF converters for non-HP gear. – Nicholas Knight Nov 9 '09 at 15:48
  • When you get into situations where you need more than 2 10gig uplinks to your provider, you'll probably have the budget for 10gig switches such as the ones from force10 (S2410) or extreme networks (x650); these things are $6,000 to $18,000 for 24 or so ports, including a bunch of sfp+ or xenpack for uplinks to your provider. That is, unless this is a university, in which case your budget for connectivity is probably zero... – chris Nov 9 '09 at 16:09
  • 1
    @chris: That's an excellent point. This is definitely a for-profit enterprise, and if we actually do get to the point of pushing 20gbps, we'll be well past break-even and the cost of adding 10GigE modules to the blades and breaking out to 10GigE switches will be of no concern whatsoever. I've been in "keep this startup alive on a shoestring"-mode for so long, it's clouding my judgement. Thanks very much for screwing my head back on. :) – Nicholas Knight Nov 9 '09 at 16:19
  • At the point where you're pushing more than 20gb of data per second, you'll want to break out the pass-through connections on your blade chassis and connect those to something like the force10 switch, which has 20 cx4 connections and 4 xenpak connections so you'd be able to connect two blade chassis together directly and have 4 uplinks. Past that you'll start to get into real money... – chris Nov 9 '09 at 16:28

Smells awfully like you're looking at the HP Virtual Connect Flex-10 interconnects (455880-Bxx) for the C7000 chassis? if so they I'd STRONGLY advise you to EITHER stick to CX4 throughout or go for the SFP+ (presumably 455883-Bxx ?) throughout. We've ran into all sorts of problems when mixing the two and if you moan to HP enough they'll seriously drop their SFP+ prices (they list about $1,319/each but I think we're buying them for about £208/each after discount).

Hope this helps, come back to me if not.

  • Nope, the blade system is from SuperMicro. Like I said, budget. :) I'm open to using ProCurve gear on the project if we need to, but if I can get away with converting the CX4 connectors to fiber, that's my first choice. The SFP+ port may go entirely unused; we could stick another 10GigE module in if necessary. – Nicholas Knight Nov 9 '09 at 12:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.