Biz & IT —

Microsoft: Windows 7 upgrade can take nearly a day (Updated x2)

A Microsoft Software Engineer has posted the results of tests the company …

While developing Windows 7, Microsoft ran many different performance tests to make sure the operating system was an improvement over its predecessors. One of the tests focused on upgrade performance: Microsoft wanted to make sure that an upgrade from Vista SP1 to Windows 7 was within a five percent threshold faster than an upgrade from Vista SP1 to Vista SP1. Microsoft gave three reasons for using the Vista SP1 to Vista SP1 upgrade as a baseline instead of Windows XP to Vista:

  • Windows XP is a vastly different operating system compared to Vista and an upgrade from Windows XP -> Vista would not be a good comparison with Vista -> Windows 7
  • Windows XP did not support 64-bit upgrades and we wanted to track 64-bit upgrade performance as well as 32-bit upgrades for Windows 7
  • Vista SP1 -> Vista SP1 is a valid upgrade path that exercises all upgrade code (this upgrade is commonly used by Product Support Services for a repair scenario)

The metric used was total upgrade time across different user profiles (with different data set sizes and number of programs installed) and different hardware profiles. According to the test results that Chris Hernandez, a Microsoft Software Engineer, posted on his blog, Redmond managed to get the Windows 7 upgrade time to be faster or equal within the five percent threshold to the Vista SP1 upgrade time. We've summarized the Windows 7 upgrade time results in the table below (check the above link for the exact hardware specs of the hardware profiles):

Data Profile Low End Hardware Mid End Hardware High End Hardware
Clean
No data and 0 applications
32-bit: 40 minutes
64-bit: 50 minutes
32-bit: 30 minutes
64-bit: 35 minutes
32-bit: 30 minutes
64-bit: 35 minutes
Medium User
70Gb of data and 20 applications
32-bit: 175 minutes
64-bit: 185 minutes
32-bit: 115 minutes
64-bit: 95 minutes
32-bit: 100 minutes
64-bit: 85 minutes
Heavy User
125Gb of data and 40 applications
32-bit: 345 minutes
64-bit: 355 minutes
32-bit: 185 minutes
64-bit: 165 minutes
32-bit: 160 minutes
64-bit: 150 minutes
Super User
650Gb of data and 40 applications
N/A 32-bit: 1220 minutes
64-bit: 675 minutes
32-bit: 610 minutes
64-bit: 480 minutes

The biggest thing that stands out about this chart is the very broad range of the upgrade time: from 30 minutes to 1,220 minutes. That second extreme is not a typo: Microsoft really did time an upgrade that took 20 hours and 20 minutes. That's with 650GB of data, 40 applications, on mid-end hardware, and during a 32-bit upgrade. We don't even want to know how long it would take if Microsoft had bothered doing the same test with low-end hardware.

The other interesting point worth noting is that the 32-bit upgrade is faster on a clean install than a 64-bit upgrade, regardless of the hardware configuration, and is faster on low-end hardware, regardless of the Data Profile. In the other six cases, the 64-bit upgrade is faster than the 32-bit ugprade.

We always recommend a clean installation over performing an upgrade installation. It's the best way to avoid lots of problems with a new operating system. Now we've got another reason to make the recommendation: in certain (extreme) cases it may just save you a ton of time, even if you have to back everything up and reinstall your applications.

Update

There's been a lot of commotion about the mention of the upgrade from "Vista SP1 to Vista SP1." This is not a typo: while it is obviously not something that is typically done, but it can be used as a repair method: the upgrade process reinstalls the operating system by replacing any corrupt and modified system files. In this case, the method was used so as to have something to compare against in the benchmarking process. The reason "Vista SP2 to Vista SP2" was not used instead is simple: Vista SP2 wasn't out yet when Windows 7 was in development.

Update 2

Hernandez commented on his own blog post explaining what we already guessed: the case scenario that took over 20 hours wasn't meant to represent regular users. Nevertheless, it's good to hear it straight from the source:

I got some questions about the data sets we used in this post, and the upgrade times associated with them. The "super user profile" is not a normal user; rather, it's the user profile that represents the extreme power-user who's working with an enormous data set (650Gb of user data) and a large number of installed applications. This user profile is not representative of what most "regular" users, who typically have a much smaller data set and would therefore experience a much, much shorter upgrade time.

The goal of this particular test was to verify that even in the most extreme scenario, upgrading to Windows 7 is significantly faster than upgrading to Vista. This kind of fundamental improvement is just one example of how we listened to our customers and took their feedback to build a great product.

Channel Ars Technica