[Beowulf] 10GbE Adapter Market

Mark Hahn hahn at mcmaster.ca
Fri Nov 15 20:03:51 PST 2013


> My preference is COTS as well, but, in all fairness, I'm rather unsure
> if COTS is more a commonplace term than SOHO or if I'm just acclimated
> to it more due to having been on the Beowulf list for so long.
>
> Let's have google decide:
> https://books.google.com/ngrams/graph?content=COTS%2CSOHO&year_start=1800&year_end=2008&corpus=15&smoothing=7&share=&direct_url=t1%3B%2CCOTS%3B%2Cc0%3B.t1%3B%2CSOHO%3B%2Cc0
>
> Looks like you're right.  The turning point for COTS vs SOHO was about
> 1995 from what I can tell.

guessing a lot of crosstalk with "south hampton" and "multiple of cot".

https://books.google.com/ngrams/graph?content=COTS_ADJ%2CSOHO_ADJ&year_start=1900&year_end=2013&corpus=15&smoothing=2&share=&direct_url=t1%3B%2CCOTS_ADJ%3B%2Cc0%3B.t1%3B%2CSOHO_ADJ%3B%2Cc0

about 10:1 in favor of COTS, with SOHO maybe a bit pre-bubble in flavor.

to me at least, they aren't very similar in meaning.  COTS refers mainly
to the property of being in volume production, almost to the point of being
fungible.  SOHO is more of a market segment - sub-enterprise but more 
serious than pure consumer (a SOHO switch might not be rackmount, but 
perhaps managable and in a metal case; a NAS might be desktop but with 
support for LDAP/AD, perhaps replication.)

I think the failure of 10G to ride COTS is interesting.  part of it is 
demand-side: most of the market just isn't gagging for it (there's no 
exponential Moore's law of consumption).  and vendors have utterly failed 
to make it cheap (or easy) enough to tempt consumers.  and IB, SAS, FC
have sucked up most of the oxygen where performance does matter...

it seems like 10G vendors also made a tactical error by perseverating
so much about power (and cable length).  yes, 10GbT requires a lot of 
processing, but there was a market that would have accepted 20W phys
that only reached 15M.  they lost early adopters to IB - more than 
just a beachhead.

so now what?  10G isn't COTS really, and doesn't seem to be getting there.
40/100G has a high-margin niche, but it's probably not big enough to make
it bigtime (esp without any help from HPC - firmly held IB territory.)

I'll throw this out there: Intel seems mostly directionless, and Thunderbolt
is a good example.

regards, mark hahn.


More information about the Beowulf mailing list