2

I have a rather simple .NET MVC app. There's not a lot of computing power required. However, we are expecting an enormous amount of traffic.

I've been using Blitz.IO to simulate concurrent users and at around 8000 concurrent users I start to notice a significant slow down, which I'd like to trigger an auto scale event. The problem is the CPU usage is only around 3%. So my bottleneck isn't the CPU.

Our current setup is three web apps across three different regions, up to 30 VMs.

I noticed in the Azure Web App section you can scale by Data in, Date Out, Http queue. I'm not entirely sure which one to use, or what each one means exactly. And I've been unable to find exactly what these metrics mean by googline around.

Jack Marchetti
  • 561
  • 1
  • 6
  • 15

1 Answers1

3

For web apps, you basically have data in/out, memory, CPU, disk queue and HTTP queue (use preview portal to see all of them). By only watching how any of these values in the monitor behave while you increase the number of concurrent users, you should be able to come up with a threshold to work with auto-scale even if you don't understand them completely. Here's a brief explanation of each:

CPU: Average CPU utilization (all VMS)

Memory: Average Memory utilization (all VMS)

Disk Queue Length: Count of pending disk operations. If your application reads/writes a lot of data from/to disk, this could be your gap. The more the queue the more your web server is waiting for disk I/O to continue.

Http Queue Length: Count of pending HTTP operations. If your application is receiving more requests than the web server can handle, this could be your gap. Usually CPU's fault, but not a rule.

Data In/Out: Amount of traffic ingress/egress (kilobytes). If your application transfers a lot of data in/out, the network bandwidth could be your gap.

Here's a very good tutorial to get you started into configuring auto-stale using preview portal.

https://azure.microsoft.com/en-us/documentation/articles/web-sites-scale/

Bruno Faria
  • 3,814
  • 1
  • 13
  • 18
  • At around 30k concurrent users we started throwing errors and had a pretty big slow down (13 sec avg. response) but the CPU was only at 9%. We had 30 web servers running. The site is relatively static, has one db query which is cached for an hour and we use a CDN for almost all js/css and images. We're using Blitz.Io to simulate the traffic. – Jack Marchetti Oct 24 '15 at 01:48