Frames, cells and QoS

It’s ironic that just before we heard about the planned merger of the ATM Forum with the MPLS and Frame Relay Alliance, we were questioning ourselves whether the difference between cells and frames really matters anymore.  So this seems an appropriate time to consider both the difference between cells and frames and where this difference does and doesn’t matter.

It’s ironic that just before we heard about the planned merger of the ATM Forum with the MPLS and Frame Relay Alliance, we were questioning ourselves whether the difference between cells and frames really matters anymore.  So this seems an appropriate time to consider the difference between cells and frames and where this difference does and doesn’t matter.

Frames and cells are both types of packets.  The difference is that all cells (within a given system) are the same length, and frames are of variable length.  For ATM, it was decided over a decade ago that cells would be 53 bytes, with 5 bytes of header and 48 bytes of payload per cell.  There’s nothing magic about this particular cell size; it’s the result of a compromise between the advocates of 32-byte payloads and those who wanted 64-byte payloads. Not particularly a bad cell size, but also not particularly good.

There are distinct advantages to using cells for transmission.  Since the cells tend to be short – relative to typical data frames – quality of service (QoS) is easier to implement because the amount of time that each individual packet (cell in this case) will occupy the transmission line is fixed.  By contrast, a large frame can occupy a transmission line for a significantly longer time. 

Additionally, because all of the packets are the same size, the switching equipment has a somewhat easier job.  Consequently, more cell traffic than frame traffic can be switched with a given amount of processing power. Many of today’s large router switching fabrics, for example, convert frames to cells for these efficiencies.

At the same time, cells come with baggage.  Data naturally occurs in frames, so it must be converted into cells through a process called segmentation and reassembly, or SAR.  Additionally, overhead is needed for each individual cell or frame.  If you break a single frame relay frame into five cells, you’ve just increased the overhead five-fold.

In another newsletter, we’ll examine these tradeoffs in light of today’s networking environment.