Microsoft SharePoint 2007 Technologies

232 10.11 Performance Monitoring

matches the expected values, many problems will go unnoticed until end

users complain.

Remember that SharePoint relies heavily on SQL being available and

performing up to standard. Therefore, it’s important to include SQL services

and databases in the monitoring and ensure that you fully follow SQL man-

agement best practices. You should, for example, set up alerts to ensure that

databases and transaction logs never run out of space and schedule various

management jobs, which keep the databases healthy and at peak perfor-

mance. Please refer to the extensive documentation in the SQL Online

Books for more information on SQL management and monitoring.

Last but not least, you should make sure that the alerts don’t simply go

to a rarely-checked e-mail inbox. If possible, you should try to put in place

an alert management plan that uses different means of communication to

ensure that critical alerts are dealt with even at 4:00am on Saturday morning,

within a reasonable time period defined (obviously) earlier in the SLA. Use

pagers and other devices, since e-mail infrastructure, just like your Share-

Point deployment, is subject to service availability.

10.10.8 Passive Analyzing

In addition to monitoring that the service is responding to user queries, you

should take steps to proactively seek possible clues for trouble down the line.

Periodically analyzing the event logs, for example for warnings and errors,

can save you time and effort, as you can deal with the issues before they

become serious cases possibly causing system downtime.

In addition to event Logs, you should analyze the IIS log files for usage

patterns to be able to better understand the user scenario and adapt the archi-

tecture to the way the users are really using the infrastructure. You might dis-

cover, for example, that users are performing more searches than you’ve

expected, and you can thus prepare well ahead of time, even before the per-

formance analysis catches the extended resource usage. SQL Server logs

should also be periodically analyzed for possible issues with index fragmenta-

tion and other database events that might cause problems in the future.

10.11 Performance Monitoring

Of course, the most important passive analysis method is system perfor-

mance monitoring, which can be set up to continuous log and alert based

on predefined thresholds. Most of the monitoring products mentioned ear-

lier also record performance information and thus can report on trends and

send alerts. In addition, Windows server ships with perfmon, which gathers

and logs data from performance counters (see Figure 10.5). Perfmon also

allows alerts to be setup with thresholds, although you must have a program

10.11 Performance Monitoring 233

Chapter 10

which does something more intelligent than logging the alert to system log

to make this feature useful.

10.11.1 Monitoring Strategy

It’s important to consider exactly what counters you should capture and how

often they need to be recorded. To establish a system baseline, for example,

you will need sampling data over several days to be able to draw conclusions.

With hundreds of counters available for both system and application compo-

nents, it doesn’t make sense to capture everything, thus allocating a lot of

space and resources for just recording the performance. Instead, it’s better to

continuously monitor key aspects of system performance, which give you

clues to what goes on and when you should pay closer attention, and add a

number of specific counters for deeper analysis.

Additionally, you should ensure that your monitoring strategy includes

sending out alerts when key thresholds are exceeded, such as available disk

space. You should also make sure that key performance counters, such as Pro-

cessor utilization and system queue, also generate alerts if their threshold is

exceeded for a period of time. Peaks will obviously occur; thus, you need to

take steps to ensure that you do not generate false alarms for peaks.

It’s important to be able to establish growth patterns and trends, which

will guide you through the production phase and ensure that any given com-

ponent is never allowed to become a bottleneck for usage growth in your

enterprise. You should be able to predict capacity requirements 6 months

Figure 10.5

Perfmon

performance

monitoring tool