Queue Solver
Queue Solver provides guidance in optimizing the settings for HBA queue depth. It enables you to maximize SAN traffic queue depth without hitting the avalanche point of backlogged requests that slows host response time.
Note
Queue Depth refers to submitted but unanswered read or write requests.
The analytic displays a graph of the inferred queue depth vs. read or write acknowledgement latency, and provides one of four recommendations about queue depth:
No change to the queue depth limit proposed.
It has been determined that the queue depth limit should be lowered. Lower it to {x}.
Based on this data set, it was determined that the queue depth limit is too low. If you feel the chosen data range represents normal operation, your system might benefit from raising the queue depth limit. Proceed with caution in incremental amounts.
Based on this chosen time range, your queue depth limit setting looks fine. You might see some gains if you lower the queue depth limit to {x}; however, the time gains might not be significant or noticeable.
The Queue Solver analytic examines network traffic for an HBA card, but does not directly read the current queue depth from the HBA BIOS.
To create a Queue Solver analytic:
Start by running a new Queue Solver from the Analytics home page by click the New Analytic or Run New button.
Click the Add button to select a host or ESX host to analyze.
Select a Host or ESX Host, then click the OK button. You can use the search box to filter the display of Hosts/ESX Hosts.
Run Queue Solver for at least two weeks, making sure to select a date range that is a good representation of your business cycle, e.g., includes month-end processing.
Subsequent runs should be made after queue depth settings have been changed.
Click the Run button.
If the host does significantly more writes than reads, click the Write-Based Recommendations radio button.
The queue solver analytic screen is redisplayed, showing the write-based recommended action to be taken on the queue depth setting. In this case, the read- and write-based recommendations are different.
You can click the percentile legend selections to show only some or all of the information.
Only the selected percentile curve is displayed.
Hover over the graph lines to display detail information.
To export a copy of the displayed graph, click the options menu (horizontal bars), and select the Export format from the displayed menu.
Understanding Queue Depth Results
Two graphs are displayed at the bottom of the page for Read and Write-based recommendations. You can toggle between the recommendations using the radio buttons.
A summary of the recommendation is displayed at the top.
The charts present the Avg Read Latency or Avg Write Latency along the y-axis, and the inferred queue depth along the x-axis. As expected, latency increases with queue depth as port utilization increases.
Hover over a box to display percentile values for the box charts.
Important
When you implement Queue Solver recommendations, examine the potential impact of the change on other devices sharing the same target storage controller, especially different HBA queue depth settings to the same controller port.
A single HBA with a significantly lower queue depth setting than the other HBAs sharing the controller can result in proportionally less work done, with a potential degradation of throughput for that host.
Therefore, if a large queue depth reduction is recommended for a single HBA, carefully consider reducing the queue depth setting for all HBAs sharing the storage port.
If a host has multiple HBA cards, the same queue depth should be specified for all the HBAs.