🚀 go-pugleaf

RetroBBS NetNews Server

Inspired by RockSolid Light RIP Retro Guy

Article View: demon.tech.pc
Article #71590

Re: Adaptec 2942 PCI Blues

#71590
From: J A F Bradbury
Date: Wed, 22 Nov 1995 00:00
53 lines
3306 bytes
Win95 actually (surprisingly!) does benefit from a Bus-Mastering SCSI 
card. Not as much as NT/OS2 but there is a distinct difference, also
confirmed by a couple of peoples benchmarks on comp.periphs.scsi.

o Re TR rate, from the platters.
  What matters is: RPM, sectors-per-track & dual-parallel-head or not.

  Take the Quantum Atlas
  7200rpm, 80-134 sectors per track, not-dual-etc-head. 512 bytes/sector.
   Max TR from Inner tracks ((7200/60) *  82 * 512)/(1024^2) = 4.9 MB/s
   Max TR from Outer tracks ((7200/60) * 137 * 512)/(1024^2) = 8.2 MB/s

  NT/OS2 happily use command-queuing etc, Win95 hasn't a clue re this.
  However Win95 does benefit from the bus-mastering side (thank gawd).

  For A/V use you will get 3.5+MB/s from an Atlas. I think there's
  a 2MB cache version which could be useful.

o There is a slight SCSI-Command-Overhead with SCSI drives, but mainly
  noticed under W4WG. Basically a 4500rpm EIDE will be slightly quicker
  than a 4500rpm SCSI *IF* data-density is the same. However, there are
  7200rpm SCSIs etc. This also ignores lower CPU usage for SCSI.

o MODE-3/4 refers to interface timings, ie, peak burst *i/f* TR speed.
  Not sustained data transfer; the only time it is, is when the data is
  already in the cache (small cache :( ) or when it's being read in for
  the 2nd time (will probably be got from the OS cache however).

  - PIO EIDE means no-bus-mastering which means 60-70% CPU usage which
    neither OS2 nor NT will thank you for, SCSI means <10% usually.

  EIDE: eg, WD730 (figs IIRC)
  4500rpm, 67-112 sectors per track, not-dual-etc-head. 512 bytes/sector.

  Max SDTR from Inner tracks ((4500/60) *  67 * 512)/(1024^2) = 2.5 MB/s
  Max SDTR from Outer tracks ((4500/60) * 112 * 512)/(1024^2) = 4.2 MB/s
  Practice: (OS overheads etc etc) User sees 1.25MB/s & much CPU use.

o Fast-SCSI-2      = 10MB/s max from all SCSI devices at once
  Fast-Wide-SCSI-2 = 20MB/s max   "       "      "      "

  Under NT 2x Atlas drives TR concurrently would max out F-S-2. A simple 
  fix is to have an NCR (40ukp) card for each h/d and use an OS like NT.

  Is EIDE better? No; without Bus-Mastering EIDE (ie, using PIO) you can
  only have 1 drive TR at the same time: so a "14MB/s" i/f is worthless 
  unless you can find an EIDE drive that manages 14MB/s.

o ftp.buslogic.com has core303.exe; this gives 3 figures for TR:
  Buffer (nonsense), Sequential (hmm) & Random (real life :( )
-- 
Julian AF Bradbury       Ergonomist; Human Factors & IT Specialist
Ergonomics & IT FAQs                    http://194.72.60.96/www/3b
Technical Details
Message-ID: <356008691wnr@xabcs.demon.co.uk>
Path: arc1-ca.pugleaf.net!unknown!not-for-mail
References: <46uha2$cis@camco.celestial.com> <9510302212.aa11276@father.sco.com> <475jsp$q66@cheyenne.iac.net> <476m8f$65d@camco.celestial.com> <4787pp$beu@bonduk.demon.co.uk> <816910679snz@longley.demon.co.uk> <816982679snz129.os2.7@corixia.demon.co.uk>

Show Full Headers
Complete Article Headers
X-Google-Language: ENGLISH,ASCII-7-bit
X-Google-Thread: 1036bc,8d388d918f4e0bdf
X-Google-Attributes: gid1036bc,public
From: J A F Bradbury <Julian@xabcs.demon.co.uk>
Subject: Re: Adaptec 2942 PCI Blues
Date: 1995/11/22
Message-ID: <356008691wnr@xabcs.demon.co.uk>
X-Deja-AN: 119882556
x-nntp-posting-host: xabcs.demon.co.uk
references: <46uha2$cis@camco.celestial.com> <9510302212.aa11276@father.sco.com> <475jsp$q66@cheyenne.iac.net> <476m8f$65d@camco.celestial.com> <4787pp$beu@bonduk.demon.co.uk> <816910679snz@longley.demon.co.uk> <816982679snz129.os2.7@corixia.demon.co.uk>
organization: ErgonITe
reply-to: Julian@xabcs.demon.co.uk
newsgroups: demon.tech.pc