🚀 go-pugleaf

: >Some do have problems with busmastering EIDE but otherwise you're just flat
: >mistaken.

I use busmastering ATA on a wide variety of Linux boxes, and have never
seen a problem.  new/big drives sustain 8-11 MB/s with small CPU overhead
(tested with 6.4G drives from Seagate or IBM, Intel-chipset boards,
bonnie or iozone on .4-6G files, ext2, usually 4K blocks, hdparm -c3d1m8)

: You've repeatedly claimed that workstations do almost entirely
: contiguous block file transfers (totally false), and that busmastering

there are indeed many applications where IO is mainly sequential, and Unixen
go to some length to ensure that it is.  clustering is common, blocksizes
larger than 1 sector are common, filesystem prefetching is common, drive
readahead is common.  there certainly are cases where IO is mainly
punctate fetches of 4-8K, but even those are more than a handful of 
sectors.  besides, those cases are often swapping, which can only
sensibly be considered a nonfatal error in a day when ram is $2/MB.

: EIDE eliminates the performance advantage of SCSI in a multitasking
: multithreading environment (it does not, in my EXPERIENCE, or in the
: experience of most others in this NG).

SCSI bigots will always believe it, no matter what.  it's simply not 
true though: I've measured it, why don't you?

in short:

	SCSI has a higher peak bandwidth, though current disks of either
	variety only consume 10-15 MB/s apiece.

	SCSI encompasses more variety and number of devices, though ATAPI
	offers increased variety to ATA buses.

	SCSI disks often have higher MTBF reliability ratings, though it's
	unclear whether the difference between 500K and 1000K hours is
	meaningful or accurate.

	SCSI disks are currently around 2x the price of roughly comparable
	ATA disks. 

	good SCSI controllers start at about $130 whereas good ATA
	controllers are built into most motherboards.

regards, mark hahn.
--
operator may differ from spokesperson.	hahn@neurocog.lrdc.pitt.edu
					http://neurocog.lrdc.pitt.edu/~hahn/