Performance Analysis Services

Do users complain that your computer systems are too slow? Are your systems under‐provisioned, or just not configured for optimum performance? Diagnosing the bottlenecks in server performance can be a complex and highly technical task. At Essential Software Solutions, our performance analysts have the tools and experience required to get your computers up to speed.

Performance Analysis – Case Study

A real‐life example will best demonstrate the benefit of performance analysis. Our client was receiving feedback from users that their Infor SyteLine applications seemed slow to respond at times throughout the day. Since most users access the system remotely, it was not immediately clear whether the issue was server‐related or network‐related.

Some real‐time observations on their SQL Server seemed to indicate a bottleneck in disk input/output.

At times, the load appeared quite light:
Performance Analysis

In the above example, response times are single‐digit milliseconds (4 – 8), overall disk activity is only 2.93%, and the disk queue is negligible (0.02).
Performance Analysis

At times the disk was working at 100% capacity, with more than 100 requests in the queue:
Performance Analysis

Response times were going into triple‐digits:
Performance Analysis

Although the evidence seemed to clearly indicate a bottleneck in disk performance, we wanted to be sure before making any recommendations. Using vendor‐specific diagnostic software, we monitored the disk performance over a 24‐hour period.

A summary of the collected data shows large peak read and write latencies, and an enormous peak queue depth of 138.7:

A graph of queue depth versus time showed very large queue values from around 7:00 – 10:30 AM, and 2:00 – 4:30 PM:
Performance Analysis

For an acceptable user experience, it is recommended that latency should be less than 20ms, and queue depth should not exceed the range of about 2 – 7.

Recorded latencies peaked at 77.4ms (read), 99.6ms (write), and recorded queue depth peaked at 138.7.

We thus confirmed that delays experienced by users were due to insufficient disk performance, and identified the specific causes of poor disk performance to be that:

1. The storage array used RAID 5, which is intrinsically slow for write performance.
2. Multiple servers were sharing the same array.
3. The array was rated for up to 1,160 IOPS, whereas demand peaked at 3,190 IOPS.

We made the following recommendations:

1. Use a RAID 10 array for better write performance.
2. Provide the SQL Server with a dedicated storage array.
3. Use Solid‐State‐Drives instead of Hard‐Disk‐Drives.

© Copyright 2013-20. | Privacy Policy
Site by Get Jacked Marketing + Lara Spence