Conquering Latency—An Incredible FEET

Current Job Listings

Network vendors are starting to provide 10GigE connections, switches and fabric, and given the exponentially increasing demand for bandwidth, enterprises will buy this equipment. Applications that can gobble up the bandwidth will soon follow-there is simply never enough bandwidth. It is becoming clear that those responsible for the applications in datacenters should also be concerned about the proximity of collaborating applications and the number of hops critical transactions take. The importance of this can be better visualized in FEET.

10Gig E means that the communication medium will transmit 10 billion bits per second, but what does a billionth of a second mean to a message? It means about 11 3/4 inches, almost a foot. A message response time of 40 milliseconds between two virtual machines, which would be considered important to achieve, translates to about 39 million feet, equaling about 7,451 miles, or about one third around the world. While this seems like plenty of speed to play with, it is important to remember how much wiring goes into a single computer (or network switch) and the time involved in message translation and protocol switching. If an enterprise has two datacenters for high availability in a city, a typical distance between them might be 5-10 miles. If highly collaborative systems have been communicating between sites to service critical transactions due to some configuration oversights, it becomes quickly apparent that feet begin to matter. To further the complexities, disaster recovery strategies often require a 100-250 mile distance between the two sites; if consolidation strategies force the DR site to become active to save costs, careful consideration needs to be given to how systems will interact over such distances.

While some mandates are irrevocably givens, such as DR site distances, there are circumstances that a datacenter team (including the application staff) can control to reduce excess lag.

* Assess how many hops a complete end-to-end transaction takes between machines; it may surprise many to know that some inter-site hopping could have crept in. There are now tools that can explore an applications entire suite of connectivity. We have seen this used and when the reports are provided the application team is often most surprised at the results.

* Know the critical applications that are especially latency sensitive, that also have high throughput, because rethinking their layout in a datacenter could save significant performance re-engineering.

* For critical applications consider redeploying the servers on the floor so that all tiers of the application are in close proximity. While this is counter to current layout strategies, it is going to become more prevalent as budgets tighten and consolidation ensues.

* The increasing use of XML related formats to transmit messages presents the problem of message bloat. Despite the use of these self describing messages, translations between XML formats often need to occur between systems. There is an emerging set of network appliances that can translate these messages at wire speed, saving server processing cycles and decreasing latency.

To save precious FEET, an incredible FEAT of rethinking the datacenter will be required.

Time and space are one fabric; in 1905 Einstein showed the relationship, 103 years later it is more relevant than ever.

Sheppard and Tony

Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.
Now read: Getting grounded in IoT