Re: Embedded Ethernet, looking for an efficient solution...

From: Richard (rh86_at_no.spam)
Date: 08/19/04 

Date: Wed, 18 Aug 2004 21:20:32 -0700

MM wrote:
> I am a newbie in this, so please bear with me... I
> was asked to estimate various options of embedding
> Ethernet into our future board level products.

A tall task for a newbie... good luck in your efforts; it's a big
subject.

> On one hand I am somewhat overwhelmed with the
> amount of information available, but on another
> hand I don't seem to be able to find answers to
> "simple" questions like the ones below:
>
> 1. What data rate is achievable in practice with
> 100BaseT assuming there are only two nodes talking
> to each other?

That's because "it depends". Like many things, there are a lot of
factors that contribute to the efficiency. Under ideal circumstances it
can be nearly 100Mbps each direction.

For example... big, small, or jumbo packets? IP or raw Ethernet? TCP
or UDP? Crossover cable or switch ports? Cut-through switch or store &
forward? Full or half duplex?

> 2. What CPU speed do I need to be able to support 2
> 100BaseT ports? In other words I am not sure
> whether some of the processors that have 2 100BaseT
> ports on-board can actually process packets quick
> enough to make efficient use of the theoretically
> available bandwidth...

Another "it depends". 2 stats specific to network processors -
packets/sec and bytes/sec. Packets/sec is more indicative of the
processor's speed, where bytes/sec is more about bus width and transfer
methods.

How much will you be doing with the packets? Are you an endpoint that
calculates checksums, buffers TCP, etc. while running both ports at line
rate, or are you a network router that shovels packets from port to port
with little processing on them?

E.g., TCP endpoints can have a lot of overhead, particularly RAM, and
especially on high-volume high-latency links (e.g., fast WAN links)
where huge buffers are needed for max performance.

At the line rates you spec, you'll be looking at PCI or on-chip
solutions. Requiring 2 on-chip Ethernets quickly narrows the playing
field. And then there's the question of whether the Ethernet controller
can perform at line rate, not to mention the TCP/IP stack you're likely
to license.

Relevant Pages

  • Re: Babysitting on iptables requested :-)
    ... Here's the list of ports that I see probed then I take the "Probe my ... this was a friendly probe; all packets were TCP SYNs - ... SYN is a packet that is used to initiate a TCP connection. ... >> between Windows machines, so without this a Windows machine in your ...
    (comp.os.linux.security)
  • Re: Help! "Martian" (packet) invasion via FiOS cablemodem
    ... the bytes 08 00 were not incorrect: the Ethernet header ... also to analyse packets captured with the tcpdump -w option. ... from Verizon or the Internet). ... My LAN is 192.x.y.z and the packet is from 169.a.b.c, so passing the broadcast along doesn't _sound_ right. ...
    (comp.os.linux.networking)
  • Re: Political Analysis of Security Products
    ... > bee collected nor has any evidence of such a backdoor ever really been ... send several packets to ports on the target system. ... be used for booth sides of the security game. ...
    (Pen-Test)
  • Re: Connecting 2 PCs to broadband with router
    ... >>> routers, has 4 Ethernet ports for networking. ... >>> used to connect any 2 devices which have Ethernet ports. ...
    (microsoft.public.windowsxp.network_web)
  • Re: Network traffic monitor app
    ... switch in the router to connect equipment together. ... So for traffic from a workstation to the internet it goes from the ... packets sent and the second the number of dud packets. ... one or more ports. ...
    (comp.sys.mac.misc)