ftth:
Thanks for the link to the article on network choices. However, the
article degenerates somewhat into a rant at when discussing 10/100/1000
Gb ethernet. I managed an engineering group of 70 staff on UNIX
workstations and clients from 1988 to the early 90's, and we used a 10
Mbit ethernet. When we ran out of bandwidth, I discovered it only had 2
Mb useful capacity, since it uses the broadcast oriented and contention
based Carrier Sense Multiple Access with Collision Detection
CSMA/CD) method. So I was never a big ethernet fan. I guess I'm one
of those people stuck in 1994 that the article dislikes so much. So I
looked for a tutorial to get me to 1997, if not 2000 (and looked at some
previous posts here) to try to find a more objective discussion (I generally
trust .edu or .org more than .com). The best I found so far, even if it is a
couple years old, is:
ece.wpi.edu
I'd still like to know if collision detection schemes 1000 times faster are
still 1/5 efficient: i.e. is 10 Gb ethernet really 2 Gb IP data? I do note that
the 'carrier extension' options of IEEE 802.3 suggest that the minimum
slot size be 512 bytes; "Figure 4: Gigabit Ethernet Performance versus
Packet Size" tells me that VOIP packets (40 bytes header, 10 bytes
payload = 50 bytes) would have a 1/10 efficiency! This amazing
technology takes 10 Gb and gives you 1 back! for the currently dominate
media (voice) of a converged network.
My understanding of the "GREAT DEBATE" (ATM. vs. FR. vs. IP. vs
Sonet) is that it is often muddled because these are often complimentary
protocols. And some are useful in the core (ATM/FR) , some in the
metro, some in building.
Basically we have today four happy layers (don't get confused with the
OSI layers, these are real layers) over FIBER in core networks today: IP
over (ATM or FR) over SONET (or SDH) over FIBER. While there
are a lot of ideas for eliminating a layer or two (IP over "thin sonet" is
one), they all trade off network management and QoS control for
bandwidth efficiency; so with the fiber glut, why bother? Even the terribly
inefficient voip packets over 10Gb ethernet might have a chance of
working, if you over dimension the network.
Gigabit ethernet is IP over MAC/PHY/PMD/FIBER, and basically says
there is no central management, do it yourself, just like ethernet (I'm
overstating this: RSVP, and QoS issues need some management- I'm just
not sure how much), its is really just a faster, extended ethernet.
If you make the debate about management, I suspect you will find that
people (SOHO market) want total freedom in the residence, but want
someone else to take care of everything outside starting at the wall of the
building. Enterprises are somewhat more complex; some want to
outsource at the office; others at the edge of the enterprise while keeping
management control of their own network.
If this is true, it seems to me the IP over ATM.... etc. has the better
chance at the core/metro for data networks, particularly if bandwidth is
free. But now I've stepped into the religious debate (war), which I
tremble to do given your bold declamatory cognomen. However, it may
be risky to invest in 10, 100, 1000 or 10^x Gb plays, particularly as the
last mile has a lot more issues than just high bandwidth access and flexible
management. |
