1000Mbs on a 66MHz bus - How do they do that?

It seems odd to me that a Gigabit NIC deal with data at a rate of 1000Mbs
yet the host bust is running at 66MHz.

How is it possible?

Could it be that the NIC is only capable of burst speeds of 1000Mbs?

Don

Don wrote:
It seems odd to me that a Gigabit NIC deal with data at a rate of
1000Mbs yet the host bust is running at 66MHz.

How is it possible?

Could it be that the NIC is only capable of burst speeds of 1000Mbs?

Don

The host bus uses parallel data transfer - 16 or 32 bits at a time.
--
LSR

Don wrote:
It seems odd to me that a Gigabit NIC deal with data at a rate of 1000Mbs
yet the host bust is running at 66MHz.

PCI sockets typically run at 33MHz (the standard allows up to 64-bit x 66MHz,
but this is rarely seen except in high-end gear). That provides a 32-bit data
path, with a theoritical transfer rate of 33M x 4bytes / second = 133MBypes/sec.
This is more than adequate to saturate the gigabit ethernet capacity. PCI-X
(64-bit, 133MHz) and PCI-e (1 bit, 2.5GHz)(per lane) can do it even easier.

How is it possible?

Could it be that the NIC is only capable of burst speeds of 1000Mbs?

You'll rarely see throughput of 1000Mbit because overhead comes in from a number
of places. Ethernet frame headers, IP headers, and propogation delays because
the other end needs to ACK each packet (delays can be minimised by properly
configured TCP stacks). Crappy protocols that limit packet size to something
way too small, etc.

Plus, of course, you have to get the data in the first place, which probably
means disk I/O with its inherent delays (mechanical and processing overheads),
and of course sharing the same PCI bus on which the CPU-NIC transfer is taking
place.

The speed quoted for ethernet refers to the available signalling bandwidth on
the network, aka 'wire rate'. Everything is signalled at that rate, but there
are gaps between packets.

Don wrote:

It seems odd to me that a Gigabit NIC deal with data at a rate of 1000Mbs
yet the host bust is running at 66MHz.

How is it possible?

Could it be that the NIC is only capable of burst speeds of 1000Mbs?

Don



If that 66 MHz host bus is the PCI bus, it transfers 32 (or, maybe, 64)
bits in parallel at 66 MHz, for a peak transfer rate of 2112 (or 4224)
Mb/s; hence, not a bottleneck for a 1000 Mb/s NIC. And other host buses
are commonly 64 (or more) bits wide as well -- no problem.

Also, 1000 Mb/s is indeed the burst speed of a 1Gb/s NIC: the rate at which
two successive bits are transmitted within a packet. There are gaps between
packets, and there is protocol overhead, so that the STR (Sustained Transfer
Rate) on a 1Gb/s ethernet is well below 1000 Mb/s, even if the end-nodes
to the transfer are blazing fast.
--
Cheers, Bob

In article ,
"Don" writes:
It seems odd to me that a Gigabit NIC deal with data at a rate of 1000Mbs
yet the host bust is running at 66MHz.

How is it possible?

Could it be that the NIC is only capable of burst speeds of 1000Mbs?

It only requires parallelisation by a factor of 16 to achieve that.
--
SAm.

"Don" wrote in message
...
It seems odd to me that a Gigabit NIC deal with data at a rate of 1000Mbs
yet the host bust is running at 66MHz.

How is it possible?

Could it be that the NIC is only capable of burst speeds of 1000Mbs?

Don

Thank you to all respondents. All very welcome and clear explanations.

I am very grateful to you all for taking the time to educate me.

Don