All communications in an application stack generally take a few microseconds. Some take a few milliseconds which results in higher latency also known as long tail latency.
How to Identify Long Tail Latency
Most of the storage array vendors use average latency to determine the response times. Also, average response time is by far the most commonly used metric in application performance management. If the response time is always consistent (transactions run at the same speed), the response time distribution is roughly bell curved which means the performance observed would represent the majority or more than half of the transactions. In real life situations this is not the case, most applications have few high response times or very heavy outliers and these result in long tails.
A long tail does not indicate several slow transactions, but few that are magnitudes slower than the normal. These outliers cannot be determined with average response times. A percentile analysis will indicate where these long tail latencies are present and its impact on applications. For example, a 99th percentile latency of 200ms indicates that every 1 in 100 requests experience 20ms of delay. If an online store has 10,000 transactions, about 100 transactions experience a 200ms delay which may not be a good end user experience.
Violin’s patented architecture delivers consistent performance which eliminates the long tail latency. In a real-world BI application running on a Violin Storage platform, a 99th percentile latency is around 750 microseconds compared to a competition’s 99th percentile value of 2.83 milliseconds. Violin Storage Platform delivers microsecond latencies 99% of the time.
For more information, check out our Oracle Solution Brief.