[HECnet] FDDI advice

[Paul_Koning at Dell.com]

On Oct 11, 2013, at 1:25 PM, Kari Uusim  ki <uusimaki at exdecfinland.org> wrote:


Thanks Paul for the detailed clarification!


On 11.10.2013 17:23, Paul_Koning at Dell.com wrote:

On Oct 11, 2013, at 2:28 AM, Kari Uusim  ki <uusimaki at exdecfinland.org> wrote:

...
Yes, that would be the one when converting between FDDI and 100BaseT Ethernet.

The best effort it can produce is about what 100BaseT can, because it is the lowest common nominator. FDDI performs better due to the larger packet size (which is usable only between FDDI nodes) and the lower overhead.

Larger packet size, yes.   Lower overhead, no.   Token systems of any kind are guaranteed to be less efficient than Ethernet, because you have to wait for the token.   At best, they will be slightly better under very high load than half duplex Ethernet, but full duplex Ethernet will outperform any token network under all conditions.

It's very clear that FDDI will outperform half duplex Ethernet, because there will not occur any collisions in FDDI.
But if we think about a FDDI-to-100BaseT (Full duplex) topology and the traffic flows with as large packets as possible, doesn't FDDI be able to carry about three times more payload in each packet than Ethernet? In that case one FDDI packet payload will need to be split into three Ethernet packets and three Ethernet packet payloads will fit into one FDDI packet.

Yes, the smaller packet size of Ethernet means the CPU does per-packet processing 3 times as often, which has a performance impact.   (Pretty small, if the CPU is fast.)   And three headers instead of one makes a difference, but only a bit over 1%.

Half duplex Ethernet slows down when there are collisions, but there is no wait to transmit whenever the link is idle.   With FDDI, there is (you wait for the token).   So at high load, half duplex Ethernet might be a little slower than FDDI, but at modest load, it will definitely be faster (lower latency).

	paul