Part of my team at the lab where I worked at the rocket ranch was once
tasked to quantify this and the first question you must ask is "Is there any
DEC hardware in the building, connected or otherwise?" What was determined
was that with DEC hardware present, even if not connected, not uncrated, not
powered on, it would impact network performance. The interest I had was in
that I was routing PCM voice comm's over the same net, and was critically
interested in how network loading would effect my throughput, hence, voice
quality due to effective packet losses thanks to delays. In the presence of
DEC equipment, the upper limit was 33 Kbytes per sec with a 10% net load,
i.e. 10% of the available bandwidth is transferred as actual raw data
packets. Raw data is data which is both input and output data, i.e. without
any additional routing or transfer-associated header, trailer, ecc, etc.
Making the DEC equipment happy apparently took more overhead than a PC-only
network, since that managed to transfer considerably more, close to 30
MB/min with large packets and considerably less with smaller packets.
Apparently, knowing that DEC protocols had to be dealt with caused a sharp
increase in overhead. The number of stations attempting to transmit have a
large influence on sustainable traffic load and the associated
acknowledgment traffic can be substantial, though it doesn't contribute much
to (net) throughput.
With five stations on my in-house LAN, all inactive other than the server
and the station to which the tape is attached during a backup on 10BaseT I
can happily feed an admittedly slow Exabyte 8200 which has a maximum
transfer bandwidth of 13.038 MB/min. Under 100 Base TX, it happily feeds an
Exabyte 8505 which seems to max out at just over 30 MB/min, with no apparent
effect on the rest of the traffic, i.e it does that even if I'm running two
backups at the same time. That same tape drive takes much longer than the
8200 on the 10BT because it is always working underflows, which means it has
to back up and retry the write once it has streamed past the end of its
data. My 8500 drives, capable of between 20 and 22 MB/min, fare about as
well as the 8500 on the 10BT, and are unaffected by the 100BTx. Hence, a
really small 10 BT seems to be able to handle between 10 and 20 MB/min with
no other traffic. If packets are small, especially with very small packets,
data<<overhead, so net throughput will be really small if you have a large
net using multilayered protocols with many stations active. Even a small
number of stations using small packets can burn much of your bandwidth
because they tend to transmit frequently.
My voice traffic with its short packets was VERY inefficient even though
there were no ack's and no error management of any kind over the LAN. IIRC,
Appletalk sends occasional traffic over an otherwise idle LAN. I don't know
whether it does this when there really is traffic.
The numbers you quote seem a mite low, but not embarassingly so. You are
configured as a single segment, are you not?
Dick
-----Original Message-----
From: Zane H. Healy <healyzh_at_aracnet.com>
To: Discussion re-collecting of classic computers
<classiccmp_at_u.washington.edu>
Date: Sunday, November 14, 1999 1:03 PM
Subject: OT: Effective Speed of 10BaseT
>I'm wondering what the effective speed of 10Mbit Ethernet is. I've got a
>DEC 3000/300LX that just isn't giveing me the kind of Xfer rates I expect,
>and I'm wondering/suspecting there is a problem with my calculations.
>Using a pair of PDF's totalling 24.6MB as my test I'm getting 8.34MB/Min,
>yet I'd expect it to be closer to 72MB/Min.
>
>The machine on the other end is a G4/450 PowerMac with 100BaseT, plugged
>into a 10/100Mbit switch, the DEC3000 is plugged into a 10/100Mbit hub
>which in turn is connected to the switch.
>
>Of course part of this could be the 21064/125 processor, and part of it
>could be the fact I'm using Appletalk.
>
> Zane
>| Zane H. Healy | UNIX Systems Adminstrator |
>| healyzh_at_aracnet.com (primary) | Linux Enthusiast |
>| healyzh_at_holonet.net (alternate) | Classic Computer Collector |
>+----------------------------------+----------------------------+
>| Empire of the Petal Throne and Traveller Role Playing, |
>| and Zane's Computer Museum. |
>| http://www.aracnet.com/~healyzh/ |
Received on Sun Nov 14 1999 - 17:58:12 GMT