Nils Goesche  <nils.goesche@anylinx.de> wrote:
+---------------
| David Bakhash <cadet@alum.mit.edu> writes:
| > TCP/IP handles datarates of 10M bps, about the same factor over
| > 56K bps as 56K bps is to 300 bps.
| 
| TCP/IP doesn't care about the data rate, and it can be much higher
| than 10M bps (ever heard of 100 MBit ethernet?).
+---------------

Or gigabit Ethernet (1000BASE-SX or 1000BASE-T)??  Our systems do 
over 650 Mbit/sec of TCP over GbE, user-mode to user-mode, with a
single TCP connection. [Over 800 Mbit/sec, if you allow "jumbo frames",
that is, 9000 byte MTUs, but that's non-standard.]

And 10 GbE is just about to pop out of the gate (already starting to
be demonstrated by a few vendors). But AFAIK there aren't any full-speed
host NIC interfaces for 10GbE yet, so for now the fastest TCP I know of
is still ours, at ~2.0 Gbit/sec (on a 6.4 Gbit/sec GSN link). [If we use
STP instead of TCP we can get ~6 Gbit/sec, user to user, single socket,
but we're talking about TCP, so I won't mention that further...]

+---------------
| Broadband ISDN (ATM) is even faster.
+---------------

Well, sure, all the way up to OC-192c/ATM, but there aren't very many
host NICs at that speed. On the other hadn, OC-12c/ATM NICs are fairly
common, and you can get ~500 Mbit/sec of TCP out of a well-tuned OC-12c/ATM
implementation, but I'm afraid GbE has already passed it by.

And at the higher speeds, people are dropping the ATM encapsulation and
just doing "packets over SONET" (POS). In fact, OC-192c/POS is even one
of the PHY/PMD options for the proposed 10GbE standard...


-Rob

-----
Rob Warnock, 31-2-510		rpw3@sgi.com
Network Engineering		http://reality.sgi.com/rpw3/
Silicon Graphics, Inc.		Phone: 650-933-1673
1600 Amphitheatre Pkwy.		PP-ASEL-IA
Mountain View, CA  94043