Peter Heywood 12/5/2012 | 3:24:49 AM
re: Intel Pushes Ethernet Backplanes Good post!

** I think folk point to the PC as an example of the benefits of standardizing subsystems, don't they?

** My impression is that a lot of vendors are now in a position where they no longer can afford to spend loads of time and money on developing telecom equipment from scratch.

On the one hand, the risk of spending all of this time and money and then the product failing to generate huge revenues is too great.

On the other hand, the big vendors disbanded their R&D teams during the recession, and the folk that were laid off are now developing subsystems for sale to their previous employers.

These R&D guys now either work for contract manufacturing organizations or specialist outfits like Denmark's Tpack A/S and Murton Consultancy & Design Ltd., (both of whom effectively develop the guts of Sonet/SDH multiservice provisioning platforms that are then sold by vendors as their own products)

All of this ATCA, ASI, RapidIO malarky is an effort to standardize these subsystems so vendors aren't locked in to a single supplier, and can benefit from competition from subsystem suppliers.

At least, that's my take on it.

ironccie 12/5/2012 | 3:24:49 AM
re: Intel Pushes Ethernet Backplanes I'm confused on this one. Someone please help me out. Is it a good or a bad idea to standardize the backplane? On paper, it looks good, but traditionally I've found that every silicone chip has defects. Say you do use this same backplane as your competition, what do you have to gain by having the same defect as them all? We would have the added benefit of complexity of too many people putting switches together sort of like automechanics and it would be harder for folks to go from job to job as in depth vendor knowledge is part of success (W. E. Deming).

No vendor does switching across their fabric like Foundry does. Nobody does it like Cisco. Nobody does it like Juniper. Nobody does it like Extreme. Their fabrics set them apart for the good or the better. If they all had the same boxes, then the best company would be the one with the most money? Cisco, Foundry, Juniper, Extreme? I'll go with that ranking.

I don't think it is as good of an idea to standardize fabrics as it would be to standardize algorithms to run over fabrics to provide QoS for oversubscribed egress ports (back pressure). Isn't it the hardware guys who are always ahead of the software guys? Perhaps we should build the perfect model of the functionality of a switch fabric from a customers viewpoint and then let the vendors determine how to achieve that in hardware. That will help Moore's law faster as they try for the O/O/O solution.

sigint 12/5/2012 | 3:24:48 AM
re: Intel Pushes Ethernet Backplanes Even Intel concedes that an Ethernet backplane isn't for everybody, however. One glaring limitation is that the speed grades jump from 1 Gbit/s to 10 Gbit/s, making multigigabit links an awkward fit for the backplane. "You would have to go to multiple lines of Gigabit [Ethernet], and that leads to added complexity," Kumar says. For those cases, Intel will be recommending ASI.

There was some talking of evolving a standard to use Fractional XAUI, i.e., use one lane of the four lane XAUI interface (2.5G, effective). It was supposed to fix the concern quoted above. I wonder what happened to that effort?
allip 12/5/2012 | 3:24:47 AM
re: Intel Pushes Ethernet Backplanes ironccie, you have an absolutely valid point when we are talking about pure communication boxes. None of the top 3-4 vendors (which you mentioned in your post) will "switch" over to standards based backplane anytime in the near future (5yrs), atleast not for their top line products.

however, as Peter pointed out in his post that the target segment is PC and as per the article Storage systems also. I am a bit skeptical about the fusion of Ethernet backplane and ASI/sRIO standards in these fields. My 2c on this is that Ethernet alone coupled with PCI-Express/sRIO will dominate this space. I donGÇÖt see ASI (due to its focus on low-mid end fabrics) blending well in this ecosystem of storage subsystem and PC domains. Besides why would one like to incorporate additional complexity/layering by blending AS and Ethernet? vendors would be better off with using Ethernet for their complete solution or use proprietary soln the box.

here's my take on the interconnectivity

it will be ALLIP in the storage domain eventually,
1. Storage- Ethernet
2. PC/Embedded device - PCI-Express
3. Wireless/DSP - sRIO
4. HPC - Infiniband.
5. Core/Metro routers - Proprietary
6. Enterprise switch - Proprietary

Most of the control plane to data plane connectivity will move towards PCI-Express from the current PCI base.

would like to hear ur thoughts...

ironccie 12/5/2012 | 3:24:44 AM
re: Intel Pushes Ethernet Backplanes Allip,

Convergence is a delicate topic... What about v6? I think Ethernet is winning, but IP has a battle to fight with itself first. How can I converge on IP when there are 2 versions of which the v6 is far superior but a big uknown?

I see HPC, media, research, and home consumer uses for non IP networks for a long time to come. This is the importance for service providers to install L2 connectivity and not get stuck in the IP way of thinking because provider are offering L2 with VPLS on MPLS (trending that way).

There are other technologies in the battle against all Ethernet/IP. Check out the work being done on sensor networks (802.15.4) where it makes no sense to have the inferior IP protocol ;-) This is being promoted by the gentleman who invented Ethernet.

As Peter points out, there is reason for them to believe they could succeed in selling standardized fabrics, but there is also reason to believe that the technology front runners will run faster.

Sign In