Databases Reference
In-Depth Information
Figure 13-1
From the perspective of a user, when a server is under an extreme load with
many outstanding queries, some of the execution time is spent waiting in the
queue, and some time is spent actually working on the query. You might com-
pare it to waiting in line for a food in a crowded baseball stadium. When the
load is high, the wait time can easily exceed the time to actually do the busi-
ness. Hardware 2 has better resources (CPUs, memory, network) as com-
pared to Hardware 1. Hence if you run the server on Hardware 2 you obvi-
ously get a better throughput since the saturation to maximum users occurs
at about 200 users.
If you have a well architected server then
Figure 13-2
shows the average
query response under load. The average response time starts to increase,
and eventually the average response time will increase linearly. If your system
only needs to support 10 to 50 members, you don't have to do anything. But if
your system needs to support 100+ users, you need to identify the system
bottlenecks. Typically servers will expose performance monitoring counters to
expose internal values. You can use the task manager and performance
counters to identify bottlenecks; whether the system is CPU-bound (such that
the Server CPU is pegged at 100%) or memory-bound (memory usage is
constantly maxed out) or disk-bound (reads or writes to disk). By removing
the system hardware bottleneck and adding more CPU or memory to the
server, you should be able to get performance improvements and support
more concurrent users as shown in
Figure 13-2
for Hardware 2.
Search WWH ::
Custom Search