A server baseline allows us to so how changes we make to our servers affect the performance of the server. it also allows us to be, dare I say “proactive” in managing our servers. Creating a baseline involves collecting data that could lead to a bottleneck. A bottleneck is a point in the path that data travels in which it has to wait. If your data is stuck, so are your users. The data the you need to collect for a baseline will vary depending on what you have on that server. This document covers just the basics.
For a basic baseline, we need to look at 4 areas: Processor, Network Interface Card (NIC), RAM, and the hard drives. We will be using the Windows Performance Monitor to collect this information for us. A baseline is taken over a period of time. Because of this, we will be using the Data Collector Sets in the Performance Monitor to do our collection.
To start the Performance Monitor, click Start, type Perfmon and press Enter
Expand Data Collector Sets.
Right click User Defined and then click New \ Data Collector Set
Provide a name for the baseline.
Click Next
Select Basic and click Finish.
Click Baseline.
Double click Performance Counter
Click Remove to remove the \Processor(*)\* entry.
We are now ready to add our basic counters to collect our performance data.
Processor
The processor is the brains of the computer. Like us humans, our brains can only do so much before it starts to slow down. We need to take a good look at the activity of the CPU to see if we are asking to much of it. When reading the data for the processor, remember that spikes to 100% are normal and should be expected. Sustained activity above 75% shows a potential bottleneck. If you have this sustained activity, take a close look at what is being done on this server and consider offloading some of the applications to another, less utilized server.
Let’s add in our counters:
Click Add
Expand Processor
Select %Processor Time
Notice in the Instances of selected object box, you may have more than 1 processor. I the example below, there are 4 processors labeled 0-3.
You may want to monitor each processor individually. Click each processor and then click Add.
Physical Disk
The Physical Disk represents each physical hard drive on your server. You may have several logical disks. That is a physical disk with several partitions. To get an accurate look at the disk usage, we need to look at the physical disk. We are interested in the %Disk Read Time and %Disk Write Time. This tells us how often our disk is being used. We also want to look at the Average Disk Queue Length. Should this number stay above 4, you may have a bottle neck. With today's high speed devices, that number can be much higher before users notice anything.
Memory
For more than a decade now, your PC has been able to use more memory than what you have installed in the computer. PCs and servers use a technology called virtual memory. WHen the physical memory is full, but the system needs to load more information into memory, the server will look at the contents of its physical memory that has not been used in a while, and write it to the hard disk. When it needs to use that content again, it copies more information from RAM to the hard disk and then grabs the information it needs off the hard disk, and stores it in RAM. RAM is fast, disk is slow. Anytime you need to do this swap, it is called a Page Fault. Because Page Faults require the use of a disk, they slow things down. A lot of Hard Page Faults/Sec indicates that you need to add more RAM to your server. In the Memory object, add the Page Faults/Sec counter.
Network Interface Card (NIC)
The next item to take a look at is the NIC. We must allow Fantasy Foot to be played without delay! Two things to look at here is how much data is waiting to travel out of the computer, and how many errors does that card receive. The Output Queue Length will tell us if there is a traffic jam trying to get out of your computer. This can indicate that your network connection is not fast enough. You may need to upgrade the NIC, the infrastructure, or both. Another option may be to add an additional NIC to the server. The Packets Received Errors will tell you if there is a bad NIC on your network.
An optional metric to monitor is in the Server object. Take a look at the Server Sessions to get an idea about how many users are using this server.
Click OK to save the counters.
Right click Baseline and select Properties.
Click Schedule tab. Here you can schedule when this performance counter starts. Under the Stop Condition tab, you can control how long the sampling takes place.
After a period of time, or after you deploy new functionality to the server, you will want to re-run this baseline and see what the effect has been. By knowing the utilization state of your servers, you can make a more informed decision about what can, and cannot be added to a servers work load.
Comments