home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.arch
- Path: sparky!uunet!iWarp.intel.com|ichips!ichips!glew
- From: glew@pdx007.intel.com (Andy Glew)
- Subject: Re: Request Info Re Bus Trends & Intel 486s
- In-Reply-To: pettengill@cvg.enet.dec.com's message of Sat, 25 Jul 92 05:31:24 GMT
- Message-ID: <GLEW.92Jul26200841@pdx007.intel.com>
- Sender: news@ichips.intel.com (News Account)
- Organization: Intel Corp., Hillsboro, Oregon
- References: <rjmartin.711191773@extro.ucc.su.OZ.AU> <1992Jul20.003506.23290@theus.rain.com>
- <GLEW.92Jul23192755@pdx007.intel.com>
- <1992Jul25.053124.14505@e2big.mko.dec.com>
- Date: Mon, 27 Jul 1992 04:08:41 GMT
- Lines: 63
-
- |>[Burkhard Neidecker-Lutz]
- |>Can you explain which *desktop* applications need peer-to-peer communication ?
- |>I.e. what is wrong with going through main memory ?
- |>>
- |>>[John Theus]
- |>>To use today's favorite buzz word: multimedia.
- |>>
- |>[Andy Glew]
- |>First you'll put video on your screen.
- |>
- |>Then you'll want special effects: things like zooming in on a
- |>particular speaker (or the papers on his desk) in a video conference,
- |>things like cropping speakers from different video inputs and
- |>superimposing them into a virtual conference room.
-
- I assume from the tone of his note that Andy is arguing for peer to peer...
-
- Not really.
-
- Peer to peer has a performance advantage (if it can be achieved).
-
- But if you want to do more processing than either of your peers can
- handle, you'll want to go to the CPU.
-
- So I'll take my typical waffling stance:
-
- (1) I will never base a computer system architecture [*] on the assumption
- that peer-to-peer communication for things like video is the only
- important thing.
-
- (2) I will try to optimize communication involving the CPU as much as
- possible.
-
- (3) I will also try not to *prevent* peer-to-peer communication from
- being used.
-
- [*] Not if the architecture is intended to stay up for several years.
- If it is a purely one-shot deal, I guess that the thing to do
- is to look at the technology wheel and try for the optimal solution
- projected for your delivery date. But the wheel turns at a rate
- that approaches 1 turn/2 years (with major cycles of 1/8),
- so if your design is going to be on the market that long plan for both.
-
- It is wise to avoid self-fulfilling prophecies. E.g. if you cripple
- your CPU by not providing the mechanisms to do fast bcopy, because you
- are assuming that extrernal hardware can always do it better, you have
- stuck yourself in a hole that it will be difficult to get out of when
- the technology wheel turns. Conversely, if you don't put the features
- to do good peer-to-peer without CPU intervention in your bus, then you
- won't be able to get the performance boost it provides.
-
- Best of both worlds?: TANSTAAFL. Emphasize the aspects that will be
- most important during your critical market window.
- --
-
- Andy Glew, glew@ichips.intel.com
- Intel Corp., M/S JF1-19, 5200 NE Elam Young Pkwy,
- Hillsboro, Oregon 97124-6497
-
- This is a private posting; it does not indicate opinions or positions
- of Intel Corp.
-
- Intel Inside (tm)
-