Transaction Queue Time

From wiki.gis.com
Jump to: navigation, search

Transaction Queue Time Sensitivity

Transaction queue time is more significant for heavier processing loads (larger transaction service times increase the probability of concurrent service transactions). Queue time is also sensitive to the number of available service providers, or processor core, available to service the transaction request (eight core server can process high volumes of service requests much more effectively than a single core server). Figure 5a-1 shows the relationship between queue time and the number of server processor core.

Figure 5a-1 Transaction Queue Time Platform Core Sensitivity

This engineering chart can be used to identify the display response time for a transaction with 1 second service time. Response time as a function of platform utilization is plotted for five different server platform configurations (1 core, 2 core, 4 core, 8 core, and 16 core). A response time comparison is made at 80 percent platform utilization (demonstrates how to use this chart) for the same 1 second display transaction. The single core server response time is 4.9 seconds; while the eight core server response time is 1.6 seconds (over three times faster display response time). This chart can be used to translate display service time to expected user response time for any transaction (multiply the results from this chart by the display transaction service time). Transactions with higher service times will have much higher response times.

This optimum platform configuration can reverse itself when deploying in a Virtual Server environment. Current testing with virtual server environments suggests multiple core machines do not scale linearly; there is a processing overhead that grows with the number of VM core. Figure 5a-2 shows a throughput comparison for a 4 core configuration (two 2 core machines vs one 4 core machine), first when there is no processing overhead (physical machines) and then with virtual machines with 10 percent overhead per processor core.

Figure 5a-2 What provides the best throughput?

The first two workflows are configured on separate physical machine environments, the first tier with two 2 core machines and the second tier with a single 4 core machine. Throughput at 2 second response time is 41,970 and 44,010 respectively; the 4 core configuration provides the higher throughput.

The second two workflows are configured on separate virtual machine environments, the third tier with two 2 core virtual machines (20 percent performance overhead) and the second tier with a single 4 core virtual machine (40 percent performance overhead). Throughput at 2 second response time is 34,970 and 31,430 respectively; the 2 core configuration provides the higher throughput.

Technology continues to improve, and all virtual machine environments today do not performance exactly the same. We expect in time the processing loads between physical and virtual platform environments will eventually approach what we see in the physical server environment. Until that time, the smaller virtual machine configurations will continue to outperform the larger ones.

[Return to 27th Edition Display Response Time]

[Return to 28th Edition Display Response Time]