[lkml]   [1997]   [Sep]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: I20 to cause problems for linux et al. (fwd)
Hi !

I2O touches a side-project of GGI (EvStack), so I might give some facts I
dug out while checking out what to do about I2O.

> The non disclosure agreement ONLY applies to the hardware info to create
> cards using the faster bus, last I checked Linux wasn't in the hardware
> biz, if it was a $5250 fee would be the least of the neccesary expences.

> What this means for Linux: $5000 is cheap enough to allow for a lot of
> clones.

> Sit back and relax all this really means is that some very fast video
> cards are comming. :)

> Please don't panic, this is NOT a bad thing.

This should be clarified a bit :

For now the I2O Bus is designed like some kind of "SCSI for IO" i.e. it switches
from a Memory/IO/Bus-like interface as we have with most todays cards to a
Command/Message oriented interface. This means, that drivers can be split
quite like Linux does for SCSI into a hardware-dependent level (for the host-
adaptor) and a hardware-independent level for handling the protocols.

As far as Linux is concerned, the point is how much of the protocols will
be revealed to the public (hopefully all) and if the hardware-vendors of
the bridge-chips give us enough information for writing the lowlevel-drivers.

The situation is quite the same as with SCSI today. The only decision that
matters to us, is if the I2O protocols will be released for public use (not
that I would doubt we would be able to reverse engineer them anyway ;-).

Another point is, that I2O as it is now, will probably not bring "some
very fast video cards". Video cards won't get faster due to having another
bus. Again I2O compares to a conventional bus like IDE to SCSI:

- The conventional bus has less overhead in getting data to/from the card,
as no additional protocols have to be handled. However
- The advanced bus (SCSI/I2O) normally compensates for that by being more
"intelligent" and handling whole requests all by itself.

Quite as you can give a complex command to a SCSI device which will handle
it all by itself, you will be able to give complex commands to I2O devices.
Thus what you will see with I2O under a good OS like Linux is the same as
what happens with SCSI: System load drops for the same operation. This
doesn't mean the operation is done faster (IDE disks and SCSI ones are
quite equally fast nowadays, too), but CPU usage drops.
In true multitasking environments this will give you more power, yes.

However AFAIK the only working I2O chipset available yet is from Intel and
designed as a (PCI ?-)bridge, i.e. commands will probably go CPU->PCI->I2O.
But this will probably get a direct one sometime.


Andreas Beck | Email : <>
=== World-Wide-Web URL : ===
======== GGI - The Right Thing To Do : ========

 \ /
  Last update: 2005-03-22 13:40    [W:0.054 / U:0.672 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site