I believe that the traditional wisdom of the industry is that it is about 65 to 70% of normal capacity. One argument for this number comes from queuing theory which suggests that when demand exceeds 70% of capacity, queue times for service increase at an explosive rate and throughput plummets.
So by retaining 70% of capacity, reasonable service can be maintained, albeit slightly degraded, and far less capable of responding to traffic surges. Significantly below 70% and heavy congestion caused by excessive queue times will occur with throughput cut to a fraction of normal.
This all points to requiring a minimum of three paths and around 33% on each path. Back at my former employer Telstra, there was a design rule that no more than 40% went on any one cable system. The slightly increased figure provided additional scope to favour a route which was more preferred either for economic reasons, superior routing, perceived greater reliability or a combination of these. But it was an imperative that no less than 60% of capacity would be retained on other facilities in the event of a cable failure.
It is not simply a question of choosing that last path just to be able to say that you have effective mesh networking. During a failure, the retained capacity must be workable. If there is excessive difference in the latency between the paths, this can have a substantial impact on the operations of some processing systems which expect data to arrive within an allowable time window to enable proper sequencing. Small variations are not an issue but large ones can be.
So the idea of having a path going say from Singapore to the US via South Africa to back up a couple of routes across the Pacific could present a problem for some of the users of the capacity. Fortunately on many of the principal routes of the Asia-Pacific, there are sufficient cables to enable effective networking although with differences in prices, there can be a tendency to fudge the percentages. In other parts of the region, new cables are being planned to address the need for the greater reliability from mesh networking.
John Hibbard is CEO at Hibbard Consulting. He can be contacted at: jhibbard@bigpond.com
