All the experiments presented so far were done over an almost idle Ethernet to ensure repeatability of our results. However, we would like to find out how the performance of remote memory paging is affected by the load of the network. That is why we repeated our experiments using an already loaded Ethernet. The results showed a performance degradation even when the Ethernet was lightly loaded. This situation is by no means surprising since the paging itself uses all the bandwidth it can get. Adding more sources of traffic leads to an enourmous demand for bandwidth causing repeated collisions and lowering the effective bandwidth of the network, leading to throughput collapse.
Fortunately, this inefficiency is not inherent to remote memory paging but rather to the CSMA/CD protocol employed by the Ethernet . This means that it is still beneficial to use remote memory paging over networks that employ other technologies (e.g. token ring), as long as they are able to provide to remote memory paging an effective bandwidth of 10 or more Mbps.