Cheap switch with a couple 100G

I keep hearing how cheap 100G is compared to 40G and it doesn't seem to hold true. Prove me wrong.

Cisco Nexus and Arista both have switches with 48x 10G ports and 2x - 6x 40G ports for under $1k. Swap those 40G for 100G and you're at $5k - $7k.

Am I missing some cheap switches with 100G?

I ask this because the transport companies seem to have given up on 40G.

The argument should be 50G is cheaper than 40G.

Because serdes is 25G typically, you get 25, 50, 100 without gearboxes
and retimers, so less pincount, less thermal, higher density, lower

But BOM impact to pricing isn't high anyhow, unless we're talking
about massive port counts.

If box DOES support 10GE and 40GE then the cost benefit is not there,
so you need to be targeting quite specific use case, use-case large
scale DCs are targeting.

Under 1K for 48 10G ports? Are you missing a decimal place?



Cisco Nexus 3064, Arista 7050sx, etc.

I don’t need 32 of them, though. 2 - 6 would be fine, 4 - 6 would be ideal.

I haven’t seen anyone selling 25G or 50G transport.

That's because, in active transport at least, 100G makes far more sense.

You may start seeing passive 25G WDM soon. Finisar have a DWDM tunable,
I believe.

That’s because the DWDM solutions (at least the good ones, wisely) use single-wavelength 100Ghz coherent light (like, you know, normal optics from the rest of history), and do not play this 4x25g lane splitting game. Unfortunately they are still quite expensive and physically large, smallest I know of right now is in CFP2.

  • Ben Cannon, AS15206

It wouldn’t be hard to do any standard wavelength, really. They just need an appropriate mux.

I'm really not sure that your statement makes sense by itself.


Are you saying that you can buy a new Cisco Nexus 3064 or Arista 7050sx for $1,000 new from these vendors, or are you talking about used stuff on eBay? If you are comparing to used stuff on ebay pricing good luck. I doubt you will find many used 100G switches as they are too new of a technology unlike 10G. Hell you can barely buy a couple of 10KM 100G optics from for less than $1000.

No, not new. No need to buy new switches when there are so many used available (except for now needing 100G). Switches have an extremely long life. I have a client that has 15 year old Foundry switches that just work, though we’re looking to replace them to get some 10G ports.

The pricing was part of my point. For years everyone says just skip 40G and go to 100G. The price difference isn’t that much… but it is.

“Everyone just skipped 40G and went for 100G.” Then why is there such an availability of 40G switches? Obviously they weren’t skipped, but purchased and then later replaced.

Looks like $280 for an LR4 40G and $800 for an LR4 100G. Still a premium for 100G over 40G.

I've seen these and their friends on the fleabay for about $1000 in working condition. It does happen, though I'd not say you can just drop in and BIN one any time any day.

I picked up a Ruckus/Arris ICX7650 with 40G module (so 2x 100G, 4x40G, 24x10G, 24x1G) for $700 a few months back on the fleabay, but that was a somewhat unusual deal, and that platform is not without its limitations (especially L3), but it's a perfectly capable box.

Of course, you're buying on fleabay, but if that's compatible with your network operations, then why not. It usually works fine. Worst comes to worst, buy two or three for less than a new one costs, and you have on-hand spares.

Whole bunch of Apples and Oranges mixed here.

* 40G in the ISP/Telco world is more or less non-existant. There was never a good uptake on any products aimed towards
ISP/Telco that did 40G cost-effectively. This is partly because 40G was/is never a thing in the DWDM-world either. There
was some STM256 built and some OTU3 built aswell but not nearly in the same density as 10G was and 100G is built today,
it had a weird timing with being fairly close to 100G and abit to distant to cost-effective 10G so it never had any good
usecase. So if you buy services from a Telco, its very likely that 40G handoff is the least preffered option for them.
Or most likely not a option at all (like from every company ive been involved with)

* You compare market-economics with what new stuff cost vs used stuff on ebay/liquidators cost. Thats not a sane way of
doing math, the reason why old 10/40 nexus and arista etc is plentiful on ebay is because people switched it out to
something newer, faster. There isn't currently anything faster on the market then 100G-switches so naturally there isn't
much of that available used. Now that at least some product-families (spine switches) getting 400G with the new
TH3-switches we can assume that there is a decent amount of older 100G spines getting decommisioned, and hence getting
available in the second hand market (maybe even late next year). However there has been _a lot_ of datacenters being
lifted from 10G/40G to 25/50/100G architecture the last year, so it makes sense that the predecessor is on ebay.

If you are to buy *new* stuff id say that going for 100G of 40G architecture makes sense in almost every aspect,
regardless of what products and usecase you have. If you rely on second-hand, then sure, 40G might be a decent choice.
However i would cry if id have to buy 40G optics today since i know they be binned in a year anyway. (4x10G PIR is still
usable optics in 100G land though).

If it is passive, you could tell them it is for 10G but use it for 25G?

Used Cisco Nexus 93180YC (oh, I guess New, sorry, it was on eBay, so I assumed used) is about $6,500, though I’ve heard they go down to $4k from time to time.

So then maybe that’s the problem… not enough used 100G gear out there yet to make it affordable… leaving 40G as the biggest cost-effective option. Most people I know don’t buy that kind of gear new, only used. It costs too much, otherwise.

From what I’ve gathered from active DWDM platform vendors, it doesn’t matter what you throw at it (up to the latest and greatest), they just eat it up (assuming you have the right line cards). Maybe it only works that way when talking to sales\marketing?

The mux isn't the problem, it's that there aren't SFP28 optics commonly
available in C/DWDM wavelengths. Yet. If they were, well maybe...

... However, your trouble then is that 25G will have similar loss
characteristics to 4x25 100GBASE, which to put it simply, isn't as
favourable as your existing 10G transceivers. You will *really* begin to
care about how 'direct' your cross-connects are.

Coherent optical transport has become far more common in recent years
for the same reasons, and pizza-box solutions for this are even coming
in whitebox guise now (see Facebook/Cumulus).

On the retail side, if you're buying 'grey' wavelength services from
optical network operators as opposed to running your own transport, they
now tend to be bundling everything into coherent line sides through the
use of muxponders. The problem with buying 25G services then becomes
"our vendor doesn't discount as hard for the 4x25G muxponder part as
they do for the 10x10G part!", or "we'll have to buy this for you
especially, and so you're footing >25% of the bill".

Chicken & egg: someone has to move first... And I don't see the ASR9k
and Juniper MX BUs rushing to support 25 & 50G.

At this point, with 400g coherent in production never mind long-haul testing; why bother lighting with anything slower than 100g coherent, especially at essentially the same price. It just makes no sense. It got skipped. We’re better for it IMO.

- Ben Cannon, AS15206

Actually FS has SFP28 CWDM optics (1270-1330) available but they are not up on the website, just as an FYI.

Single Wavelength Coherent or 4x10g coherent?
- Ben Cannon, AS15206