Midrange 16-20 port Ethernet Switch Design

[ajb]

I have an application for a set of custom ethernet switches--the application requires a rugged form factor and some additional dedicated signalling to be integrated, hence a custom solution.  Solutions for ten or fewer 100Mb ports are easy enough, and cover part of our application space, but we also have some application areas where something with 16-20 100Mb ports and 2-4 1G ports would be more suited.  At that port count, the parts I'm seeing get a LOT more complex, and I'm leery of the amount of development work that would be required to bring the product to market.  So I have a 9-port solution in progress, and I'm currently scoping out options for the higher complexity application.  Our quantities wouldn't be huge, so recouping development costs is a concern.  Our application would ideally include basic traffic prioritization and VLAN tagging, but we don't need any really intricate management features.  It would be nice, but not essential, to have the option of a fiber interface (using an off-the-shelf transceiver), depending on cost/complexity that that adds.

Anyone have any experience developing mid-range ethernet switches (16-20 100M/1G ports) willing to share your experience? 

How complex is it to bring up something like a Sparx III/IV?  Obviously the hardware is nontrivial--BGAs and DDR3--but that's at least a known quantity, and can be contracted out if necessary.  What's less apparent to me at the moment is the level of software development required.  Is the integrated processor that seems to be typical in that class of parts involved in all traffic handling, or is it possible get a basic switch with minimal management up and running without much CPU intervention?  Does the available software support from the manufacturer(s) typically make it fairly easy to get a more complex managed system up and running? 

Lastly, any other manufacturers I should be looking at?
- Micrel-Microchip: Max out at 9 ports, limited selection
- Microsemi: seem to have a good range of products and reasonable availability in low quantities.
- Broadcom: Don't seem to have much in the ~24 port range, poor availability
- Realtek: have a few interesting options, but don't seem to be readily available

[AndyC_772]

IIRC, Realtek are distributed by Future, so you should be able to get their parts if you need to.

Have you looked at Marvell semiconductor?

[ajb]

Thanks, forgot to mention Marvell.  As far as I can see through all of their marketing fluff, there might be a part or two in the Prestera line that would be reasonable fits, everything else seems to be higher performance or lower port count.  Apparently Avnet distribute them, but they don't seem to be catalog items.

[awallin]

The White Rabbit switch is an 18-port 1Gbit SyncE switch, open hardware + software:
https://www.ohwr.org/projects/white-rabbit/wiki/switch

there's a lot of components in that design for precision timing - you wouldn't need them for just a 1gbit data switch.
There should be fpga firmware for the switch part in that project - might be useful if you are serious about your own switch...

[ajb]

The White Rabbit switch is an 18-port 1Gbit SyncE switch, open hardware + software:
https://www.ohwr.org/projects/white-rabbit/wiki/switch

there's a lot of components in that design for precision timing - you wouldn't need them for just a 1gbit data switch.
There should be fpga firmware for the switch part in that project - might be useful if you are serious about your own switch...

That's a neat looking project, but an FPGA based solution looks like a ton of extra work, including adapting the all-SFP design to mostly copper PHYs, and in the end I have to spend an extra ~$2k on silicon per unit.  So that's not a great value proposition versus an application-specific SoC that I can get for ~$60 including--at least according to the brochures--a decent application-specific software stack to build on.  Shaving $2k off the BOM pretty quickly adds up to a good chunk of NRE!

1. Contact a professional switch developing company, like TPLink, and ask for a special design. This will likely to have an MOQ, but should be way less than IC manufacturer MOQ.

There's actually a manufacturer that's already building switches for our market that are half way to what we want, and they do OEM work as well, so partnering with them is an option I already have in mind.  Even if we go that route, though, I'd like to have a better handle on the development work that goes into such a project so I can be better equipped to have that sort of conversation.

2. Clone an existing similar Realtek-based design and beef up signal integrity and power supply.

I'm not aware of any reference designs for the class of switch I'm looking at (although there may be some behind manufacturer NDAs).  I've seen designs for <10 port devices, but those are easy by comparison.  I could crack open a few commercial devices, but that only gives me an overall view of the hardware, which is already more of a known quantity, and no real insight into the software development, which is my larger concern.

NDA with Microsemi is in progress, so we'll see what happens there.  Just getting access to the datasheets for some of their parts will be a big help!