Don't forget the backplane of the switch: just because says it a 10/100 or gig does not mean you can push that much data across the backplane. Worse yet is when you end up with three or four gig connections into a crapola switch and, in that case, you might get 100 meg out of it sustained (and you'll be lucky if you get that). When I was doing the projects for the Army we would design to have 10 meg to the desktop sustained. It doesn't sound like much until you account for the total number of network devices and multiple that by 10 meg. Your core components have to be very, very fast and have very, very low latency.

I am finishing up a project where I'm placing 10 gig router interconnections are the core. It is awesome.

Greg


On Dec 9, 2004, at 10:55 AM, Aaron S. Joyner wrote:

Brian Henning wrote:

----- Original Message ----- From: "Christopher L Merrill" <[EMAIL PROTECTED]>


"Gigabit Ethernet: This term does not connote an actual operating speed
of 1GB/sec. For high-speed transmission, connection to a Gigabit Ethernet
server and network infrastructure is required."


What does that mean? Does that mean I need something besides a switch
with a gigabit port?

I think the description was trying to explain the maxim that a chain is only as strong as its weakest link... There are plenty of consumers (/idiots) out there that would not immediately grasp why they only get 10Mbit through their 10Mbit hub to their 10Mbit server when their workstation has 1Gbit. Dell's just trying (vainly, probably) to point out that you'll only get gigabit performance if all the steps along the way are gigabit.


Cheers,
~B

Brian does a good job of pointing out the actual purpose of Dell's statement, but there's a greater subtlety that bears mentioning. If you use cheap hubs, and regular PCI Gigabit NICs, you won't see 1,000 Mbits / second. In order to get up into that range, you're going to need to overcome two bottlenecks.

First the PCI bus can't really handle that much throughput, so you're going to need to use something like Intel's CSA architecture (which ties the Gig-E controller into the north bridge chipset, bypassing the PCI bus), or go straight to PCI-X, preferably on a well-designed motherboard using AMD's HyperTransport.

Second, you'll also need to up the MTU of the interface and use what are commonly referred to as "Jumbo Frames", meaning that you send data in packets larger than the standard 1500 byte chunks. The maximum practical MTU for Gig-E is usually 9000 bytes, but take note that most inexpensive Gig-E switches won't support the larger frame sizes (and will drop or truncate those packets).

Hopefully this will help someone who's looking into Gig-E understand how to set their expectations accordingly. :)

Aaron S. Joyner
--
TriLUG mailing list : http://www.trilug.org/mailman/listinfo/trilug
TriLUG Organizational FAQ : http://trilug.org/faq/
TriLUG Member Services FAQ : http://members.trilug.org/services_faq/
TriLUG PGP Keyring : http://trilug.org/~chrish/trilug.asc



-- TriLUG mailing list : http://www.trilug.org/mailman/listinfo/trilug TriLUG Organizational FAQ : http://trilug.org/faq/ TriLUG Member Services FAQ : http://members.trilug.org/services_faq/ TriLUG PGP Keyring : http://trilug.org/~chrish/trilug.asc

Reply via email to