[c-nsp] Cheap 10G between 7600 and Procurve 5406zl

[Lincoln Dale]

On 17/03/2010, at 7:05 PM, Marian Ďurkovič wrote:

> On Wed, Mar 17, 2010 at 09:54:13AM +1100, Lincoln Dale wrote:
>> from a switch design standpoint if you are designing a switch that could
>> be used in many places in the network then reality is one probably needs
>> to support multiple transceiver types if you want to address all 
>> requirements.  nothing new here.
> 
> This is in fact very new and unwelcome development. By now, all tranceiver
> types (be it GBICs, SFPs, XENPAKs, X2s or XFPs) were able to support whole
> range of applications albeit not always from day one.

the evolution of transceiver types has always had implications.  just perhaps you haven't had visibility in to them.
you assertion that "all applications" have been available in all types is not correct.

from XAUI to XFI has always had implications on transceiver types.
for example having 4 lanes of 3.125Gbit/s makes XAUI based transceivers more suitable for LX4 and CX4.
conversely, having a single serial transmission of 10.3125Gbit/s with XFI makes those transceivers more suitable for CX1.

could those permutations be done?  absolutely.  but only in a suboptimal manner.
to do LX4 or CX4 in SFP+ would requires turning 1x10.3125Gbps into 4x3.125Gbps.
not impossible.  but most definitely a higher 'cost' to do so versus it "coming for free" in X2 which is XAUI based.

1000baseT could not be done in GBIC format originally due to power consumption.
same holds true for DWDM in SFP+ today.


> And, as mentioned
> before, they always had deterministic host-to-module interface which ensured
> compatibility and exchangeability.

this assertion is also false.  i can categorically state that there has not been, there have been any number of quirks with "standards compliant" MSA transceivers.

your experience may well have been "plug and play". part of why that is the case is exactly because of testing/qualification.

in either case the EDC issue to what i talked about is really only specific to LRM and Passive CX1 on SFP+.
SR/LR are just fine.

> 
> Datacentres certainly do have their specific requirements, but this hardly
> justifies the recent hype and rush for SFP+ form factor. In fact, another
> vendor was already bitten by this - they released SFP+ based linecards first
> but based on customer feedback the same linecards appeared with XFP slots
> several months later.

i can't speak for other vendors' choices, but dare i say it if they blindly transitioned to SFP+ in a market where the requirement was different than predominantly SR/LR optics then that is probably lack of clue on said vendors' part.

reality for those platforms at Cisco that use SFP+ is that they _are_ predominantly targeted at datacenter, where CX1 offers significant price advantages over 'fiber' transceivers and we can build port density into modules and switches that could not be achieved with either X2 or XFP.  furthermore where SFP/SFP+ can be interchanged it means the same port can be used for 1G SFP or 10G SFP+.

physically a XFP is 71mm x 18.5mm x 8.5mm.  SFP/SFP+ is 56.5mm x 13.4mm x 8.5mm.  shoe-horning up to 48 x 10G ports into 1RU would not be possible with XFP.

again - maybe those aren't the characteristics YOU are looking for in a product.


> 
> And as far as I heard, linecard with X2 slots is also planned for Nexus 7000

its a bit more than planned. http://www.cisco.com/en/US/prod/collateral/switches/ps9441/ps9402/data_sheet_c78-574915.html


> - although it's beyond my understanding why Cisco pushes this old XAUI-based
> bulky form factor into the latest switch types...

what is used behind the scenes (XAUI, XFI) has no bearing on you as an end customer AFAIK - beyond whether something is available or not.

> Or is XFP form factor
> still banned on Cisco switching plaforms for business reasons ? 

i'm not aware of any "ban".


cheers,

lincoln.