[c-nsp] SFP+ modules in Catalyst switches (was: Nexus 7000)

[Richard A Steenbergen]

On Tue, Jan 29, 2008 at 09:22:53AM +0100, Marian ??urkovi?? wrote:
> On Mon, Jan 28, 2008 at 11:07:13PM -0500, David Prall wrote:
> > It uses SFP+'s, they supposedly will be available in both 1GE and 10GE.
> 
> While the move to SFP+ for the Nexus7000 is clearly the only solution
> (datacenter needs high density & low cost 10GE links and Nexus has
> the potential for 500 Gbps per slot), the SFP+ datasheet says it will
> also be adopted in Cat6000/4000/3000 series.
> 
> Could someone please explain the motivation for this?
> 
> >From the customer's perspective, it's a nightmare to see fourth different
> 10GE form factor coming into e.g. Cat6500, especially when XFPs were
> already used on 10GE SPAs and ES20 cards. Consolidation on XFP form
> factor would be IMHO much better for several reasons:
> 
> - it's also XFI-based
> - all types are available *now* (including CX4, ZR and DWDM which are not
>   planned for SFP+)
> - supports both LAN and WAN PHY
> - offers sufficient port density (24) for 80 Gbps per slot backplane
> - has the widest adoption rate as it's used in SONET/SDH, DWDM and ethernet
>   equipment
> - is subject to continuos inovation (tunable XFPs, ultra long reach XFPs
>   etc.)

Don't forget, SFP+ is limited to such low power (1.5W in the highest power 
class) that it is all but impossible to ever power long reach optics (40km 
or 80km+) with them.

The motivating factor behind SFP+ is the ability to do high port density 
and crazy oversubscription, since SFP+ is so much lower power than XFP. 
Note that if you compare the same types of optics (i.e. SR to SR, LR to 
LR), the power difference between SFP+ and XFP is minimal (yes you're 
offloading a CDR component, but you're just moving it off to the host 
board so in the end you stay pretty neutral). The difference is that SFP+ 
has no higher power classes (2.5W or 3.5W as in the case of XFP long 
reach), therefore it will never be possible to put a long reach optic in 
one, therefore you can "safely" design a 36 or 48 port card for them.

This is fine if all you're doing is datacenter or small campus stuff, but 
completely destroys the capability to do long reach/DWDM optics over dark 
fiber. Personally I suspect that SFP+ is not going to be particularly 
popular with the SP crowd, and any vendor who values their business should 
make a high-density (16-port or so) XFP blade alternative as well.

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)