NetNews Usenet Archive 1992 #31/NN_1992_31.iso/spool/comp/sys/next/advocacy/3458

Internet Message Format | 1992-12-24 | 3.9 KB

Path: sparky!uunet!overload!dillon
From: [e-mail] (Matthew Dillon)
Newsgroups: comp.sys.next.advocacy
Subject: Re:  100 Mips Intel NeXT (moved from comp.sys.next.misc)
Distribution: world
Message-ID: <[e-mail]>
References:  <[e-mail]> <[e-mail]> <[e-mail]> <ijeff.724698606@cunews>
Date: 23 Dec 92 12:12:31 PST
Organization: Not an Organization
Lines: 75

In article <ijeff.724698606@cunews> [e-mail] (Ian Jefferson) writes:
>In <[e-mail]> [e-mail] (Matthew Dillon) writes:
>
>>In article <[e-mail]> [e-mail] (Scott Byer) writes:
>>>Matthew Dillon writes
>>>
>>>>     Furthermore, The EISA bus specification has no problem with
>>>>     all elements required for a workstation... DMA, vectored
>>>>     interrupts, etc...  Since I'm not familar with how much of the
>>>>     specification has been implemented, I can't remark on existing
>>>>     EISA machines, but I don't see the problem with EISA.  (ISA,
>>>>     on the otherhand, has huge problems with DMA).
>>>
>>>Current EISA cards tend not to be able to run at the speeds required.
>>>Also, the 16bit wide bus limitation is pretty severe.    (Are there
>>>32 bit versions available yet?)
>
>>    Main memory is not on the EISA (or ISA) bus and hasn't been for years.
>>    Video is not even on the bus anymore.
>
>>    The only thing that's happenning on the bus, in fact, is I/O, and
>>    considering the fact that the NeXT is generally memory-bound I/O is not
>>    really a problem.  Our (admittedly high end) PCs get 3MB/sec through
>>    the filesystem, about 3 times what a NeXTStation get.
>
>Wait a sec ...
>
>What kind of PC's running what OS have you got.  I've never seen DOS machine
>move files as fast as a NeXT can do ethernet. I've worked on some 66Mhz DELL
>machines even.
>
>I'd really like to know.  I have a small simple C benchmark that I run that
>writes a large file (50 or 100 Mb) and I get about 700 to 1Mb/sec tranfer
>rate to disk on a NeXT.  Even the fastest I/O machine that I know of the SGI
>Crimson or 4D35 only gets at best 1.7 to 2.3 Mbs on really fast (not striped)
>disks.
>
>I think you are seeing the effects of cache on your PC, if you increase the buffers on
>a NeXT, or increase RAM,  you will get about the same effect.

    Ok, I got the goods... it was a DPT SCSI controller on an EISA
    platform. It did not get the kind of throughput I thought -- only
    1MB/sec through the filesystem.  We were moving 30MB files around which
    easily defeats any caching the thing might have had.  We also had a
    SCSI II controller (don't know the manufacturer) that was getting
    2MB/sec through the file system with the same test.

    However, in talking to our PC expert he noted that many machines these
    days have several (up to 3) LocalBus slots which essentially go at the
    CPU's bandwidth.  Performance wise, you put a video card in one, A
    network card in a second, and a hard drive controller in the third and
    poof, done.

>If you want to try a very simple example copy a 10Mb file structure like shakespear
>to /tmp on a  NeXT. Then do the same thing with your PC's.  I bet your NeXT will
>outperform any PC about 3:1, at least that has been my experiance.

    Unfortunately it doesn't... I only get 1MB/sec through my NeXT. (A mono
    turbo 040 slab).  Whatever the NeXT is, one thing it is not is fast. I
    get better performance from my Amiga with a 25MHz 030.

                    -Matt

>--
>---------------------------------------------------------------------------
>Ian Jefferson    [e-mail] No NeXT mail please!
>        [e-mail] NeXT mail please!

--

    Matthew Dillon        [e-mail]
    1005 Apollo Way        uunet.uu.net!overload!dillon
    Incline Village, NV. 89451    ham: KC6LVW (no mail drop)
    USA             Sandel-Avery Engineering [phone]