Which is more efficient?


In your humble opinion, which transmission method is more efficient, packet or cell? Granted a cell is a fixed length packet and an IP packet is variable length....would this necessarily only relate to a specific protocol, namely, cell in ATM, and IP in Ethernet or other types of domains....feedback highly welcomed. Trying to make a decision on the transport mode for cost, delay, jitter, ROI, etcetera.

Jay Murphy
IP Network Specialist
NM Department of Health
ITSD - IP Network Operations
Santa Fé, New México 87502
Bus. Ph.: 505.827.2851

"We move the information that moves your world."

Confidentiality Notice: This e-mail, including all attachments is for the sole use of the intended recipient(s) and may contain confidential and privileged information. Any unauthorized review, use, disclosure or distribution is prohibited unless specifically provided under the New Mexico Inspection of Public Records Act. If you are not the intended recipient, please contact the sender and destroy all copies of this message. -- This email has been scanned by the Sybari - Antigen Email System.

Packets can have a max size as well based on the path MTU, such as 1500
bytes in an Ethernet (10/100) link. I think there are a lot of other
variables here such as are you billed per data unit, bandwidth and control
factors on the links, and what type of data is being sent.

If your data can always fit in a smaller N-byte cell, that can be quite
efficient since you have minimal overhead or wasted space and all the
benefits of the fixed length data unit from a processing standpoint.

If you are constantly fragmenting and then having to reassemble data due
to the small cell size, you would be better off with a variable length
packet, especially when bandwidth is less in demand than processing power.


What type of traffic are you looking at sending?

As Scott said smaller payloads that need to be sent quickly work out well in
fixed cells but larger payloads would be better off in variable sized

Also are you looking at simple data transmission rates or are you wanting to
factor in hardware load, backplane load, cpu efficiency etc?

When you say "transmission method" are you just interested in packet/cell forwarding, or are you also including the effort involved in segmentation and reassembly?

And when you say "efficient" are you talking about power consumption, or cost per bit, or payload versus header, or minimising jitter for isochronous applications, or something else?

If the question is a pragmatic one (e.g. "which will allow me to get the most sleep, and spend the least money") then perhaps at low speeds, with Nortel's bankruptcy imminent, you could expect to find a lot of cheap ATM gear on the used market that would be the right short-term answer. It'd have to be pretty cheap though. I have met clueful people who have come to this conclusion, astonishing though it seemed to me at the time.

At higher speeds, you might find that ATM gear either doesn't exist, or is so ludicrously expensive compared to ethernet switches that it makes you laugh coffee all over your keyboard.


In my humble opinion, if you care about actual in-the-field efficiency as
opposed to theoretical or in-the-lab results, I think you'll find that there is
enough statistical spread between "best" and "worst" actual hardware for both
packet and cell to swamp the theoretical benefits - there are good packet
processors out there that will kick the butt of most cell gear, and there's
good cell gear that will outperform some packet gear.

And then there's cost issues - if cell is 5% more efficient, but 35% more
expensive, is it really a good choice? (Unless of course you *need* that 5%
to fit through a non-negotiable bandwidth notch someplace - but then you'll
be screwed *anyhow* if your traffic grows 7%).

It really depends on what your applications are.
I've spent the last decade as the regional ATM specialist (among other
things) for an
international carrier, and since we can sell you koolaid in ATM,
Frame, MPLS, VPLS, IPv4, and IPSEC flavors,
I can be fairly neutral about technology recommendations for my customers.

The most efficient transmission method is the one for which you know
how to set up your router
to match the way the carrier's network works, so you'll need to train
your people.
If that's ATM, you may need to do some ATM-specific things, and
they're different for different carriers;
if it's Ethernet, you'll need to decide how to handle access line
failure detection.
And the work you need to do is much different if the
ATM/Frame/Ethernet is a Layer 2 end-to-end service
or if it's an access line for a routed service such as MPLS.
ATM can give you really good control over jitter, but only if you set
it up correctly.
Dedicated Ethernet access typically has lower jitter than shared
switched Ethernet access,
but it only comes in a couple of sizes and may cost more if that's
bigger than what you need.

As far as cost-effectiveness goes, ATM cells have about 10% overhead,
but some carriers price their services to charge you for it and some don't,
and they have different policies about bursting;
what you really care about is what price they're going to charge you for the
data circuits you need.
Ethernet also has a lot of overhead, if you're carrying lots of small packets;
it's very significant if you're carrying VOIP, and trivial on big file

These days circuit costs have decreased enough that router costs can be a
significant part of your total cost. ATM cards are traditionally expensive,
but if you're buying a VLAN-based switched ethernet access service,
ask your router vendor what size router you'll need to handle traffic shaping -
even if the Ethernet is built-in, a large teal-colored box costs more
than a medium box.

My main concerns about ATM, other than whether it matches your applications,
are whether it'll scale to the size you need, and how long you'll be able to get
good router vendor support. I don't see Frame/ATM interworking going away
as a method for handling lots of small endpoints like cash machines reliably,
at least until there are good ways to manage thousands of IPSEC sessions,
but it's not the technology you're going to want for OC48s.
DSL is usually ATM underneath, but that may or may not be how you
connect to your DSL carrier.