For a few weeks a while back, people in my house had been bugging me to fix our Internet problem. From time-to-time, devices were not able to connect. No eMail. No NetFlix. No Minecraft (this last one was devastating).
I tracked the issue down to my wireless router. The 4-5 year old device was extremely hot. There were 50+ devices connected to it! When I’d originally purchased the thing, I probably only had 10 devices connected.
After purchasing a new one, all of the connectivity issues went away.
As I started looking at the features of the new router, I saw that there is a traffic monitor built-in. So, I turned it on.
Here’s the thing I found–while looking at it at 3am–interesting with this graph:
There is a spike in activity in the middle of the night. I tracked this over a few days and the spike shows up consistently. After doing some digging, it turns out that these spikes are my Windows 10 PCs and my Windows 7 Media Center PC doing their nightly updates.
This brings about an interesting point with regards to Windows 10 optimization. Although we said in Windows 10 Optimization Results blog that the optimizations gave us about a 20% boost in server density, what we have to remember is that many of the optimizations we implemented won’t be accounted for in the test time period.
Take the scheduled task “Customer Experience Improvement Program/Consolidator” as an example. It runs every 6 hours starting at midnight. In order for this optimization to be reflected in our test results, the test must be run at one of the respective intervals. If my test only runs for 1 hour, there is a high probability I will not have a test running when this task is executing.
Many of the other scheduled tasks run at startup. Most performance tests I’ve seen only focus on the steady state, which means most of the startup scheduled tasks are also missed as part of the test.
So, does this mean our 20% benefit for running the optimizations are false? Of course not, but it does indicate that over the course of a workday or workweek, the benefit might be larger than 20%.
But in the end, nothing will ever be better than a real-world comparison.