How busy can I run the CPU? - Windows Server Capacity Management 101(7 of 12)

How hard you can work a CPU is highly dependent on the type of CPU and the type of work it is doing, there is no one size fits all number for how hard to work them.

Newer = more capable - Newer CPUs have larger on-chip cache memory, allowing more instructions to be kept nearer the cores. Cache memory is quicker to access than main memory but there is much less of it Megabytes of very fast cache, Gigabytes of fast ram and Terabytes of slow disk.

It’s also not just about the speed of the clock – a 3.4 GHz Pentium IV is nowhere near as capable (as in “can get through work”) as a brand new 2 GHz Xeon processor. Because the newer CPU can do more in one clock cycle then a older CPU can do in many.

More cores = can be pushed harder - It’s all about THROUGHPUT, not just speed.

The more cores a CPU has the harder you can run it without performance problems.

Dependent on the type of work; hyper-threading or not…

Some best practice recommendations are:

1 core

·      At 50% busy will take twice as long as if it were 0% busy.

·      At 80% busy it will take 5 times as long.

·      At 90% it will take 10 times as long. 

You can see with one core it does not take much work to make throughput slowdown.

 2 cores

·      At 50% it will take 1.3 times as long.

·      At 80% it will take 2.7 times as long.

·      At 90% it will take 5.2 times as long. 

16 Cores

•       50% - 1x, 80% - 1.02x, 90% - 1.22x

For a 16 core CPU it does not make much of a difference running it at 80% compared to 50%. But you have to keep in mind that all configurations max out at 100% just the number of cores flattens out the curve. You can see with more cores it takes longer to hit the knee of the curve.

Benefits of Multiple Cores

This chart illustrates that as you add more and more cores the response time curve is more and more flat. 

On Thursday I'll be looking at how to monitor and manage CPU, don't forget to sign up for our next workshop 'IT Capacity Planning' http://www.metron-athene.com/services/online-workshops/index.html

Josh Worth

Consultant

 

Best practice recommendations - Windows Server Capacity Management 101(6 of 12)

So now we have gone over what we need to properly manage a windows environment, here are some best practise recommendations.

There are 3 main components to monitor in your windows systems:

·      CPU – physical utilization

·      Memory - usage

·      Disk – occupancy and performance

These are all components that if they fill up or are over utilized will severely effect performance.

Best practice recommendations - CPU 

What to monitor 

The first component to look at is CPU. When monitoring CPU you need to understand the difference between Logical CPU and Physical CPU, if your system is virtualized then it will be logical CPU as the windows environment does not know about the physical CPU it is being hosted on. 

·      If physical, CPU Total utilization of the machine - a physical system is much simpler as you are directly monitoring the physical components. 

·      If virtualized, CPU usage by the guest system - you will need to know the Physical CPU usage which is under the hypervisor. If you only look at CPU busy and it says 80%, it could be 80% busy of the 5% that has been allocated to it by VMware. You need to look at process level CPU busy. 

·      Process-level CPU busy; if virtualized gives a view of relative usage of the physical CPU busy from the host. It shows you how much CPU time each process is using, this is useful to see where all your CPU time is being used.

On Monday I'll be looking at how hard you can work a CPU.

Josh Worth

Consultant

Balance service against cost - Windows Server Capacity Management 101(5 of 12)

The better service the more the cost - When it comes to balancing the cost of a service it is important to know what the impact of spending too little or too much will be. 

Align your IT spend to your business needs – It’s not about spending more and more it’s about spending smarter. Understanding what the business needs are and understanding how to meet them in a cost effective way.

It’s important to know the wider business, if you are expanding by 50% you need to know what is needed to meet the new demand. Without forward-looking activities, you could be in for any number of unpleasant surprises, such as:

·      Performance crises

·      Unnecessary hardware expenditure

·      User dissatisfaction.

Understanding how to meet demand - Capacity Management is responsible for ensuring adequate capacity is available at all times to meet the requirements of the business.  It is directly related to the business requirements and is not simply about the performance of the system’s components, individually or collectively.

On Friday I'll be discussing best practice. Don't forget to sign up for our IT Capacity Planning workshop running in September.

http://www.metron-athene.com/services/online-workshops/index.html

Josh Worth

Consultant

Trends - Windows Server Capacity Management 101 (4 of 12)

The purpose of trending is predicting what will happen by what has happened, the accuracy of trends rely on what is happening will carry on happening into the future.

Importance of trends – A trend gives you warning if your demand is going to outstrip then your supply and gives you chance to act.

How long to trend forward - As with most things there is no one size fits all. When deciding on the length of trends it is important to take a few things into consideration such as how long it takes you to buy and install new hardware. There is no point in trending forward on disk space for a week if it takes you 2 months to get additional space.

So a good length of a trend is how long it takes you to procure, physically install and configure new hardware, if this take 3 month then that is how long your trends should be.  

Trending is good at predicting when something will hit a threshold but not telling you what will happen when it does, this is where modeling comes in.

·     
Importance of modeling - it allows you to see how a system will react under different workloads. If the business has an event coming up that means it servers are going to be under higher than normal load you want to be able to reassure people that the system can handle it.

·      What modeling shows - Modeling will show you how your components will perform under different workloads, and what component will fail and when.

Modeling is used frequently for ‘what if’ scenarios such as “What if my workload increases 30% will my system handle the extra load or will it fail? If it does where will it fail?”

Knowing this lets you be proactive instead of reactive.

On Wednesday I'll be looking at Service versus Cost. Talking of Wednesday we've got a great webinar lined up for you 'Capacity Management from the ground-up' ..don't miss it! http://www.metron-athene.com/services/webinars/index.html

Josh Worth

Consultant