Specifically, I'd love to see a case study comparing Sitara's
proprietary transport protocol to an HTTP/1.1 transaction over TCP that
supports Slow Start, Congestion Avoidance, Fast Retransmit, and Fast
Recovery Algorithms a la RFC 2001
<http://www.internic.net/rfc/rfc2001.txt> and Selective Acknowledgment
(SACK) a la RFC 2018 <http://www.internic.net/rfc/rfc2018.txt>.
As way of reference, Windows 95 and NT have atrocious TCP stacks,
supporting only Slow Start and Congestion Avoidance, which has been
required under the Host Requirements
<http://www.internic.net/rfc/1122.txt> Section 4.2.2.15 since 1989.
Fast Retransmit and Recovery have been around since 1990 (more brilliant
work by Van Jacobson), while SACK only reached consensus in 1996.
Anyway, the betas of Win98 and NT 5.0 support all of these TCP
performance enhancements, which together do what Metcalfe seems to say
Sitara does.
If Sitara does have additional optimizations, are they willing to turn
change control over to the IETF? I don't mean to sound so critical,
Keith, but I think this is a very tough niche to survive in. I'd
recommend that they build up a lot of hype over the coming months and
then sell out to an OS vendor (Sun or Microsoft).
BTW, the mention of Fall Internet World on http://www.sitara.net makes
the site seem dated.
- dan
P.S. Why would I possibly bother to learn about TCP enhancements?
Researching <http://www.teledesic.com/tech/latency.html>.
> -----Original Message-----
> From: rohit@godzilla.ics.uci.edu [SMTP:rohit@godzilla.ics.uci.edu]
> Sent: Thursday, December 11, 1997 1:31 AM
> To: FoRK@xent.ics.uci.edu
> Cc: connolly@w3.org; khudairi@w3.org; jg@w3.org
> Subject: Sitara's SpeedServer Technology Debut a la Metcalfe...
>
>
> Bob Metcalfe's column from a week ago heralds the debut of Sitara,
> a company we all know Keith is affiliated with. This stuff is hot
> shit.
> And I think they have a pretty wise standards strategy, too :-)
>
> Rohit Khare (Live from the IETF where we just shut down the HTTPWG
> meeting series after three years...)
> ======================================================================
> ======
>
> [From the Ether]
>
> December 1, 1997
>
> Follow the shining star's lead for a better, faster Internet
> -- but don't tell
>
> You have to promise, before reading any further, that you
> will not mention this column to the Internet Engineering
> Task Force (IETF; at http://www.ietf.org) or the World Wide
> Web Consortium (W3C; at http://www.w3.org).
>
> I have to insist, not because I fear upsetting the IETF and
> W3C -- that's my job -- but because a new company, Sitara
> Networks, is this week introducing products that might make
> IETF and W3C look bad. Sitara asked me to be sensitive,
> because it is a member of both and wants to stay friends.
> So, promise?
>
> Sitara was founded in 1996 by Malik Khan, formerly of
> Motorola. He's since gathered a team from places like Bay
> Networks and Lotus, raised $12.5 million in venture capital,
> and moved to Waltham, Mass. "Sitara" means "shining star" in
> Urdu, the language of Pakistan, which is where Khan is from.
> Find all of this at http://www.sitara.net.
>
> Sitara will this week introduce SpeedSeeker and SpeedServer.
> The company's slogan is "the wait is over," by which it
> means the World Wide Wait, one of my favorite subjects.
> Sitara is promising potential Web-server customers that
> their users will experience a Web that is three- to
> eight-times faster. And the busier the Internet, the greater
> the speedup.
>
> So, how will Sitara speed the Web?
>
> What a SpeedSeeker-equipped Web browser does with a
> SpeedServer at a Sitara customer's Web site is to control
> and optimize transmissions through Internet links between
> them. Sitara combines HTTP and TCP, and makes them work
> better together.
>
> HTTP and TCP handshaking are combined and streamlined.
> Element requests from processes on the client are
> consolidated. Transmission rates are controlled to the
> capacity of limiting the bottleneck between client and
> server, often a modem. When receive buffers threaten to
> overflow, transmissions are slowed before packets are lost.
> And when backbones grow congested, transmissions are
> controlled accordingly. In particular, when packets are
> lost, only the lost ones are retransmitted.
>
> The benefits of Sitara's HTTP/TCP optimization are fewer
> dropped connections, less data transferred, lighter server
> loads, and quicker responses, especially as Internet packet
> losses mount during busy hours.
>
> Sitara shows a speedup of 800 percent with backbone packet
> losses running 15 percent. Losses of 50 percent are often
> recorded, such as when the stock market recently collapsed
> -- oops, wrong word.
>
> If SpeedSeekers and SpeedServers accelerate Web access as
> Sitara claims, IETF and W3C will look bad for two reasons:
> their processes for improving HTTP and TCP are slow, and
> they're organizationally reluctant to combine HTTP and TCP.
> So, shush.
>
> Now, some of what Sitara is doing sounds like HTTP/1.1,
> about which I was hopeful here once. (See "Replace the
> 'net's old pipes with shiny HTTP 1.1 for few floods, fast
> downloads," March 31.)
>
> Jim Gettys, a Digital visiting scientist at W3C who has
> never heard of Sitara, says HTTP/1.1 will "probably go to
> draft standard sometime in the next couple months, as we
> wrap up interoperability testing and documentation." (See
> http://www.w3.org/protocols/activity.)
>
> On the subject of HTTP/1.1 deployment, Gettys says, "We
> still have a small fraction of clients talking to a
> significant fraction of servers, resulting in an even
> smaller fraction of traffic being HTTP/1.1. This will change
> as browser deployment hits its stride."
>
> Sitara says its three- to eight-times speedup is an
> improvement over browser and server implementations already
> using HTTP/1.1. But of course, such implementations are by
> most accounts not very good yet.
>
> Egads, is Sitara proposing to deviate from the standard HTTP
> and TCP? Not exactly. First, a SpeedSeeker-equipped browser
> (Microsoft or Netscape on Windows 95 or Windows NT)
> encountering a non-SpeedServer site just lets HTTP and TCP
> thrash around as usual. And vice versa. But when a
> SpeedSeeker client detects it's talking to a SpeedServer
> (Unix or NT), Sitara's "Turbo TCP" kicks in.
>
> Also, SpeedSeeker runs atop the standard IP and presents
> standard interfaces to applications so that they need not be
> modified.
>
> Further, Sitara, hoping to mollify IETF and W3C, promises to
> submit its protocol technology for standardization through
> IETF and W3C.
>
> If you use Windows 95 or NT, take 5 minutes to download the
> 1MB beta version of SpeedSeeker.
>
> Let me know how true Sitara's claims of speedup are for you
> in this beta version. But, again, shush. Don't let the IETF
> and W3C in on this.