Subject: Re: VAX 6460 being slow, IO bottlenecks and SMP woes ...
To: None <>
From: Eric Smith <>
List: port-vax
Date: 03/21/2002 10:02:13
>> > It wasn't. It does thing that today's desktops can't do. When is the
>> > last time you got 99.999% uptime out of a PC?
>> Not counting an eight-hour power failure, one of my PC-based servers
>> has had under two minutes of unscheduled downtime in seven years of
>> 24x7 operation.  That's better than 99.9999%.
> But the point is that you cannot count on it. If PC hardware works for
> seven years without fail, that is definitely the exception and not the
> rule. PC hardware is generally mediocre. Even my Amiga hardware screams
> quality next to PC hardware.

Who said anything about counting on it?  The question was "when is the
time you got 99.999% uptime out of a PC", and the answer is "for the last
seven years straight".

I've never seen a VAX do better.  Most of my experiences with them
were actually less reliable.

But I didn't say anything that the average experience with a PC would
be that good.

>> I don't know what systems those were, but my first 486 PCI system
>> routinely got sustained PCI throughput of over 15 MB/second, between
>> disk, ethernet, and display.
> The key word above: sustained. Sustained throughput is the amount of
> data continuously moving through the bus. Video obviously doesn't count
> (we are talking about servers here, aren't we?)

Doesn't matter whether we're talking about servers or not, the fact is
that the claim that PCs could only manage sustained 1-2 MB/s over PCI
was false even in the 486 days.

>> Even many ISA bus systems could sustain over 4 MB/second.
> Let's see... 16 bits, 2 clocks per bus cycle, 8 MHz... that makes for a
> theoretical speed of 8 MB/sec. That's the theoretical maximum. Because
> of the piss poor design of the ISA bus (it was poor even in the '80s),
> the CPU cannot do anything else while bus transfers are taking place
> (64k DMA is next to useless), meaning that any system, even in a
> perfect world, that was doing 4 MB/sec would have exactly half of all
> CPU time be wasted before any real work is done.
> So, tell me - what, if any, 486 system ever existed that could do some
> meaningful work while transferring 4 MB/sec and operating with half of
> its CPU?

It is NOT the case that the CPU was starved by the ISA bus on all
such systems.  Many systems had cache and local bus.  On such systems,
I saw actual *measured* sustained bus bandwidth of over 4 MB/s on
some systems.  I'm not just making up numbers, I was involved in
performance analysis and optimization.

> Again, the key word is sustain. The context is a discussion about
> servers, so we're not talking about a synthetic benchmark and saying,
> "Yes, this ISA SCSI controller can sustain 4 meg/sec"; we're talking
> about what a server can sustain in the course of its work.

What kind of server?  I'm talking about what the hardware can handle
under a real workload, which may have little or nothing to do with
your arbitrary concept of a server.

> People like VAX because it is kick-ass hardware. It is engineering to
> an extreme. VAX are reliable, and PCs are not (yes, that's a
> generalisation).

I don't think you'll find anyone who disagrees.

> If you want to compare PCs and VAX, yet you wish to
> ignore those things that make the VAX unique by dismissing them with
> one-off examples, then you're in the wrong place.

Who was doing that?  Certainly not me.

But even so, there are many applications (even some servers) where
a PC is the right thing, and not a VAX (or current equivalent).  You
have to do a cost/benefit analysis.  For some things the VAX (or
current equivalent) wins, for others the PC wins.  Depends on the
application requirements.