Memory

One of the most important aspects of a switch is the memory. Switches are often presented with
interfaces running at different speeds. In fact, the differences are commonly factors of 10 (10/100/
1000 Ethernet). Combined with this possible bandwidth mismatch between interfaces, the fact that
switches move frames from one interface to another at very high speed means that buffer space can
fill up very quickly. The result is that the science of data buffering is quite advanced, and the simple
serial shift-register memory of the past is no longer suitable.
The reason for using fixed-size buffers in the first place is not necessarily intuitive. You might
think that better use would be made of shared memory by just placing arriving frames/packets into
the next free space and making an entry in a table, rather like the way your hard drive manages
files. But the problems that arise from this are in fact very similar to the hard drive file storage
mechanism. In short, how do we use space that has been released after data has been forwarded
from memory?
Obviously, the space made available after a packet has left the memory block is likely to
be the wrong size to exactly fit the next occupant. If the next packet is too small, space will be
wasted. If it is too large, it won’t fit, and we would need to fragment it. After a while, throughput
would slow down and more and more packets would have to be chopped up for storage and
reassembled for transmission. On our hard drive, we’d have to defragment our disk regularly.
In shared memory, we’d just end up with smaller and smaller memory spaces, with the resulting
loss of throughput.
Fixed size buffers allow us to control the way that memory is allocated.