Latency - Return of the "L" word
Industry Commentary By Frank Dzubeck ,
Network World , 02/28/2008
What is the "L" word? It's latency.
For the first time in years, CIOs are becoming concerned that networking and inter-process latency is affecting IT performance.
The new IT initiatives of the 21st century are based on the business process transformation within a service-oriented architecture — Web services, Web 2.0, virtualization, federation and information management are creating new levels of IT performance requirements.
Add organizational and supply chain transformation through VoIP, video-based collaboration and innovative real-time, industry-specific applications and we have a major festering problem.
In the past, the solution has been the same — give IT more resources and it will solve the problem. Increase bandwidth if it is a networking issue, add memory or processors if it is a compute problem, or add faster/denser disk arrays if it is a storage problem. Expensive, but this technique has always transparently solved the problem.
From a networking perspective, the issue is becoming critical in the data center. Virtualization can create massive economic benefits, such as increased utilization of physical infrastructure, reduction in power consumption, increased productivity through ease of use and so on. What began with compute and storage virtualization has extended into applications, workflow, security and management software. Suddenly the need to monitor and manage latency is becoming evident.
The era of widely distributed data centers is over. Corporations are consolidating into fewer data centers. Within those data centers, consolidation of compute, storage and networking resources will be aggressive. The classic concepts of data center construction are being rethought and now revolve around a design philosophy similar to large Internet software services providers such as Google and Yahoo.
Simplify the infrastructure and software services around business/application-purpose components. Numerous names exist for this aggregation concept. IBM has coined the latest term for these IT entities, calling them "ensembles." Taken to the extreme, one can collapse within the corporate data center thousands of servers into a shared virtual multi- processor server and more than a tetrabyte of disk storage into collocated memory-based storage — all within a single IBM z10 super-server. Latency within this configuration is reduced to microseconds vs. milliseconds and resiliency is increased 1,000-fold.
Are we back to the era of the mainframe ? No — numerous data center infrastructure design philosophies will exist, dependent upon economics, business applications, and management and governance requirements of the corporation. Clusters of compute and storage resources will always need to be networked together as a platform for corporate applications and information. The type of networking and the amount of intelligence within the network will be become evident when application, virtualization and federation latency demands and corporate governance requirements are taken into consideration.
The entire IT industry has been in denial for years over the latency issue. One only has analyze the most recent data center announcements from Cisco, Juniper and even IBM and look at the marketing obfuscation to avoid using the "L" word. This seven-letter word scares the vendor as well as the customer.
Data centers cannot exist alone. They must connect to corporate users, corporate branch offices, sales channels, suppliers and other internal or external data centers. Add to that connections to grids, clouds and the Internet. The next major networking latency issue is the WAN. Again we enter a state of denial over the "L" word when we believe that WAN acceleration is only an economic issue. The fantastic growth in the WAN acceleration industry will not cease but will escalate with the ever greater use of XML, Web Services, Web 2.0, HD Video, real-time transactions and specific application intelligence.
The service provider should not be left out of this discussion. The "L" word is so heinous in that it almost never appears in a carrier’s vocabulary. All carriers primarily sell and focus their marketing on bandwidth. That bandwidth comes with varying degrees of service-level agreement (SLA) commitment. Within that SLA are usually loose guarantees of latency measured in such a manner as to confuse the customer and to give carrier support organizations maximum leeway to absolve themselves of a specific customer problem. The carriers are learning that as they create more complex services the "L" word is becoming an even greater issue.
Hidden in the closet for years, latency in its various forms is emerging as a subtle yet complex and critical component to both business, corporate and IT success. The IT industry, from the business process concept to the business application implementation, within the data center, across the corporation and its external food chain through the WAN/Internet, must accept the negative reality of the "L" word.
IT management must begin to establish robust latency SLA policies and learn to govern, monitor and manage latency as any other critical technical issue. Do not avoid the issue — next-generation, intelligent latency-sensitive SLAs are needed now, rather than in the future.