Benchmark Wars (was Re: gauging interest in VAX 6000-530)

From: Eric Smith <eric_at_brouhaha.com>
Date: Mon Oct 25 02:30:22 1999

Chuck McManis <cmcmanis_at_mcmanis.com> wrote:
> As an empirical note the 16450 UART often could not be serviced fast enough
> by the 486DX66 to prevent it from dropping incoming characters above 9600
> baud. That was even with the "FastCOM" driver that came with Procomm plus
> for DOS, not until the 16550 added a 16 (or was it 32) character FIFO did
> it work. Now multiply that by 12? Say 9600 baud, every types a key. That is
> 12 interrupts to service while you're potentially trying to push out 12
> streams of bytes.

Which is why no one with any sense put 12 normal serial ports on a PC.
Instead there was a thriving business in multiport cards from companies
like Digi.

Which isn't really dissimilar to the situation on minicomputers. In
principle, you *could* put a whole bunch of DL11 async ports on a PDP-11.
But noone in their right mind did so. Instead, they used DH11 and DZ11
muxes.

So this is just another red herring in the attempt to prop up the
assertion that old VAXen are "better" than late 1980s PCs.

> Anyway, the VAX (and most DEC gear) got around this by having channel
> processors _everywhere_ and of course the modern PC has channel processors
> _nowhere_. (IDE taking this to the logical extreme)

Actually, almost all PCs now *do* have the equivalent of a channel for the
disk. It's not as functional as a 360/370 channel, but it's roughly
comparable to the Massbus and DSI interfaces used on VAXen. It's split
between the bus-mastering ATA interfaces of the Intel and Via chipsets,
and the local intelligence of the disk drive.

And almost all high-end Ethernet and SCSI cards have functionality
equivalent to a channel.

The 486DX2/66 systems Mike C. was comparing to obviously didn't have the PCI
bus-mastering ATA disk interface. But for server applications those
systems were typically configured with an Adaptec 1542 SCSI controller, which
served as a very intelligent channel processor.

> Somewhere I've got an 8 port serial card that was built for the 16 bit ISA
> bus (it has 8 mini-din connectors on the card edge). The thought was to run
> UNIX on a 486 and use terminals for multiple students. It didn't work. The
> PC couldn't keep up. I fully admit though that given source code I might
> have been able to make it work :-)

Yes, I've been involved with businesses that used 486-class machines to
server as many as 64 terminals for point of sale and inventory applications.

No, there was no way that the system would have had adequate performance if
it had to support that many users doing serious software development.

But then, neither could the VAX. The 8650 which I used in the mid 80s would
serve 50 users easily as long as only a few people were doing memory and CPU
intensive things. But if more than three or four people tried to do a compile
at once, performance went to hell. Which would have been OK if only the
compute-bound jobs suffered. But the scheduler couldn't cope will enough, so
even the "light" jobs became unresponsive. This is because everything that
was learned in the 1960s and 1970s about building good timesharing systems was
subsequently forgotten in the early 1980s.
Received on Mon Oct 25 1999 - 02:30:22 BST