Will 'Velobahn' Deliver?
kaizen asks: "How much processor time is lost to the lower levels of the protocol stack and not running your code? According to Akamba, by offloading the Transport and Network layers to their PCI cards, your server can get a x4 boost to throughput and x20 in response time. Check out the press release about Velobahn on ZDnet. How well does a starting price of $3,000USD compare with other options?"
At what point does this become absurd? "How much of your processor time is spent running code that could easily be delegated to an auxiliary chip?" We could put a bunch of stuff onto cards, and then the problem would be waiting for all these concurrent processors to talk to each other and get their data back to the main processor in time.
No thanks. One chip works just fine for me.
---
---
Remember when "Truth, Justice, & the American Way" wasn't contradictory?
Solutions like this may have a market, but it is an extremely limited one. If you have to ask whether to use it, you shouldn't.
This product sounds like a specialized version of
the SiliconTCP Product that InterProphet has
been developing for a couple years now. Bill and
Lynne Jolitz of 386BSD fame started InterProphet.
Their model was a full TCP/IP stack in silison
on an ethernet card. These folks sound like
they've gone a bit further up the OSI layer and
actually put HTTP requests together.
It's an interesting idea (Interprophet.com had
some benchmark showing their stuff to use 1/10
the processing power) but I personally have my
doubts about an http-specific accelerator.
I've not heard much out of Interprophet lately. Any ideas what the Jolitzes and crew are up to?
--chuck
3Com and Intel already make TCP/IP "aware" adapters - some with onboard hardware encryptor chips as well.
I want to delete my account but Slashdot doesn't allow it.