* Clusters, grids and jumbo frames
As server clusters and grids grow into the newest wave of computing, the way we think about local-area networking will change. One example of this is in the technology of jumbo frames.
In the past, there has been much resistance to jumbo frames – at least publicly. Jumbo frames are Ethernet frames that are larger than the 1,518-byte size spelled out in the standards. The fact that they are not standard is a huge deterrent for many users.
Proponents argue that when you are moving large files, you incur a heck of a lot of overhead if you limit your frame sizes – servers that are sending and receiving the frames have to do work hard to split the files into many small pieces and then reassemble them at the other end. As speeds increase, so does the amount of time dealing with this overhead. If you could use fewer pieces, the end devices wouldn’t have to work so hard.
About a year and a half ago I asked readers what they thought of jumbo frames. I received some pretty negative responses. Readers pointed out that networks are converging, so you don’t want large packets getting in the way of small voice packets that need to get through the network in a hurry. And there are other ways of dealing with the overhead.
But in the case of clusters and grids, you may want to use methods like jumbo frames to speed communications among servers that are part of these larger collectives. Depending on your network, interference with time-dependent voice traffic may not be an issue, and the priority may shift toward ensuring that the servers in the collectives have the most efficient communications possible. Jumbo frames might be the answer.
