Could older Ethernet/ARCNet NICs be overclocked successfully?
Really silly question but something I always wondered about... Back in the days of 386 and 486 I overclocked plenty of CPUs by just changing the oscillator.. I always wondered if that was also possible with network cards back in the day. I think old cards (such as LANtastic compatible cards) had a 20 MHz Oscillators to match their 10 mbps clock speeds. If you replaced this with a 25 MHz oscillator - is it possible it would have worked at the higher data speed? Is there something about the protocol or signalling that would have prevented it from working reliably, assuming the components had enough margin to handle the speed? (Note, I'm assuming you have two PCs connected, and overclock each PC's NIC to the same frequency. I'm also assuming a short enough cable can deal with higher frequencies. Also 16-bit ISA busses have ~ 64 mbps of bandwidth so I don't see a bottleneck there). Thanks! John
On Mon, Jul 4, 2022 at 6:11 PM John Heritage via vcf-midatlantic < vcf-midatlantic@lists.vcfed.org> wrote:
Really silly question but something I always wondered about...
Back in the days of 386 and 486 I overclocked plenty of CPUs by just changing the oscillator..
I always wondered if that was also possible with network cards back in the day. I think old cards (such as LANtastic compatible cards) had a 20 MHz Oscillators to match their 10 mbps clock speeds. If you replaced this with a 25 MHz oscillator - is it possible it would have worked at the higher data speed? Is there something about the protocol or signalling that would have prevented it from working reliably, assuming the components had enough margin to handle the speed?
iirc, Ethernet has a 64-byte minimum packet size to ensure that 2 packets originating at the same time but at opposite ends of the network, will collide in a way each station will notice (i.e. each transceiver will see the first few bytes from the other station, before it has finished sending 64 bytes). So, if you shorten the length of each bit by using a faster oscillator, you either have to increase the minimum packet size, or reduce the network size so that you are still guaranteed to observe a collision in that worst-case scenario. In your hypothetical situation, reducing the maximum segment lengths so that the network is 25% smaller should compensate for packets taking 25% less time to transmit, enabling things to work without changing the protocol to require larger minimum packet lengths. -ken
Ahh ok interesting. So that packet size was chosen for whatever the maximum length of Ethernet was at the time? Whether coax or cat V cable I guess ? Sounds like as long as you're 25% below the max you're probably ok ? (Or certainly ok with a few vintage PCs a few feet away from each other) Thanks On Tue, Jul 5, 2022, 3:08 PM Kenneth Gober <kgober@gmail.com> wrote:
On Mon, Jul 4, 2022 at 6:11 PM John Heritage via vcf-midatlantic < vcf-midatlantic@lists.vcfed.org> wrote:
Really silly question but something I always wondered about...
Back in the days of 386 and 486 I overclocked plenty of CPUs by just changing the oscillator..
I always wondered if that was also possible with network cards back in the day. I think old cards (such as LANtastic compatible cards) had a 20 MHz Oscillators to match their 10 mbps clock speeds. If you replaced this with a 25 MHz oscillator - is it possible it would have worked at the higher data speed? Is there something about the protocol or signalling that would have prevented it from working reliably, assuming the components had enough margin to handle the speed?
iirc, Ethernet has a 64-byte minimum packet size to ensure that 2 packets originating at the same time but at opposite ends of the network, will collide in a way each station will notice (i.e. each transceiver will see the first few bytes from the other station, before it has finished sending 64 bytes). So, if you shorten the length of each bit by using a faster oscillator, you either have to increase the minimum packet size, or reduce the network size so that you are still guaranteed to observe a collision in that worst-case scenario.
In your hypothetical situation, reducing the maximum segment lengths so that the network is 25% smaller should compensate for packets taking 25% less time to transmit, enabling things to work without changing the protocol to require larger minimum packet lengths.
-ken
On 7/5/22 15:11, John Heritage via vcf-midatlantic wrote:
Ahh ok interesting. So that packet size was chosen for whatever the maximum length of Ethernet was at the time? Whether coax or cat V cable I guess ?
Yes, it was chosen based on the maximum length for the 10base5 specification, RG-8 coaxial cable, which is 500m. -Dave -- Dave McGuire, AK4HZ New Kensington, PA
participants (3)
-
Dave McGuire -
John Heritage -
Kenneth Gober