*BSD News Article 4850


Return to BSD News archive

Xref: sserve comp.periphs.scsi:6985 comp.unix.bsd:4898
Newsgroups: comp.periphs.scsi,comp.unix.bsd
Path: sserve!manuel!munnari.oz.au!uunet!unislc!martin
From: martin@unislc.uucp (Martin Cryer)
Subject: Re: [386BSD] doesn't find harddisk: 1542B vs. 1742A (standard mode) ?
References: <1992Sep10.062637.11604@autelca.ascom.ch> <1992Sep10.195442.22178@cco.caltech.edu>
Message-ID: <1992Sep11.005422.22534@unislc.uucp>
Reply-To: martin@unislc.UUCP (Martin Cryer,D1Z01,5754)
Organization: Unisys, Salt Lake City
Date: Fri, 11 Sep 1992 00:54:22 GMT
Keywords: 386BSD Adaptec 1542B 1742A
Lines: 58

In article <1992Sep10.195442.22178@cco.caltech.edu> morphy@cco.caltech.edu (Jones Maxime Murphy) writes:
>nbladt@autelca.ascom.ch (Norbert Bladt) writes:
>
>Yes, I really see *very* little performance enhancement in enhanced mode.

Well to be fair to Adaptec, a large amount of the performance is down to
the efficiency of the drivers. The 1542 has been around for
a while and the drivers have gotten fairly fast by now. The 1740 is newer
and a lot of drivers appear to be in the "I got it running" category
(though this is not intended to imply anything about the qual of the
386BSD drivers). Reliability is desirable over speed in this
area for newish drivers I think!

Without drivers taking advantage of the 1740s architecture and
features to the full, the only real bonus is in the decreased load on the
system processor and the I/O bus hog time for a given workload. However,
this will be mainly apparent when multitasking is really happening in anger to 
more than one or two drives. Even a decent PC with a reasonably
affordable single drive will not saturate the 1542 or SCSI bus in
normal use. The 1740 though can be pushed further with faster
PCs and multiple faster peripherals. So the 1740 may appear to be
little faster than the 1542 in these restricted circumstances.

Our U6000/15/35 Adaptec 1542 / 1740 sdiscsi driver has now about the same
degree of maturity between the 1542 handling and the 1740 enhanced mode
handling portions. For things like AIM-III or TCP-A benchmarks or real world
loaded systems on multiple drives, the increase in performance of the system
as a whole is VERY noticeable. The 1740 is easily worth the cost delta.

The difference is also especially noticeable on SVR4 where effectively the
buffer cache is the VM page pool. If you have > 16MB of memory, the 1542
handler must copy any buffers above 16MB to below 16MB before the
controller can Bus Master DMA the data (the 1542 can only address
below 16MB as it is an ISA 16bit Bus Master controller). With the
full 32 bit DMA of the 1740, it can use buffers/pages anywhere in memory.
On SVR3, the buffer cache was (statically) allocated from < 16MB
physical addresses and therefore did not have the problem (unless you
were clever and did direct I/O sometimes). Obviously SVR4 with its SunOS
derived VM has advantages over SVR3, but the penalty is there for
ISA PC's with ISA Bus Master DMA disk controllers and > 16MB of RAM.

As an aside, the 1740 / 1740A have been very reliable over the nearly
2 years we have used them. Now to really fly, try multiple 1740s
(if someone else is paying!).

>Yeah, same here....
>Jones
>-- 
>Jones M Murphy Jr                                        (212) 602-4375
>NatWest Investment Bancorp, National Westminster Bank PLC
>100 Wall Street, New York, NY10005                   FAX (212) 269-0113

Martin Cryer
UNISYS Salt Lake City
SVR4.0/MP Development
A NatWest Bank (UK) customer - be careful with my money please :-) 

All opinions my own and not those of my employer....