Article View: comp.sys.next.hardware
Article #593Re: Terrible SCSI Transfer
From: David George
Date: Mon, 31 Aug 1992 11:23
Date: Mon, 31 Aug 1992 11:23
16 lines
1481 bytes
1481 bytes
In article <1992Aug29.011817.8627@leland.Stanford.EDU>, m@crito.stanford.edu (M Carling) writes: |> In article <1992Aug28.162416.164165@lexmark.com> songer@lexmark.com |> (Christopher Songer) writes: |> > |> > OK, so NeXT's Disk I/O bites.... Does anyone have a feel for |> > whether this is due to their SCSI hardware or the software |> > driving it? Thanks! |> |> My guess is that it is due to poor benchmarks. NeXT uses old NCR 5390 chips for doing SCSI. These generate a lot of interrupts for each SCSI transaction (>2). The NeXT would benchmark worse against machines using the NCR 537x0 chips or even machines with intelligent host adapaters built around the 5390 (for example). Interrupts are context switches and require pushing of registers and are therefore slow in cpu terms (== bad). The 5390 doesn't support SCSI2 and is limited in some other areas. The software does a pretty good job under these circumstances. Hopefully they will move to the NCR53720 for the new hardware. David.
Technical Details
Message-ID:
<1992Aug31.112327.2138@osf.org>Path:
arc1-ca.pugleaf.net!sparky!uunet!think.com!paperboy.osf.org!davidReferences:
<1992Aug28.162416.164165@lexmark.com> <1992Aug29.011817.8627@leland.Stanford.EDU>Show Full Headers
Complete Article Headers
Newsgroups: comp.sys.next.hardware Path: arc1-ca.pugleaf.net!sparky!uunet!think.com!paperboy.osf.org!david From: david@postman.gr.osf.org (David George) Subject: Re: Terrible SCSI Transfer Message-ID: <1992Aug31.112327.2138@osf.org> Sender: news@osf.org (USENET News System) Organization: OSF RI Grenoble References: <1992Aug28.162416.164165@lexmark.com> <1992Aug29.011817.8627@leland.Stanford.EDU> Distribution: usa Date: Mon, 31 Aug 1992 11:23:27 GMT Lines: 16