Stressed Out Windows: Making Windows Robust

What happens within Microsoft when a Windows Vista or Longhorn Server daily build is ready? Thousands of machines spin up an insane stress test that sucks up all available memory and handles and then pushes the machine to the limits with an intensive set of feature stress routines.

This is an incredibly tough test that helps us find obscure race condition bugs and assess the overall robustness of the operating system.

Join Dr. Sneath and Stress Technical Lead, Eric Hansen, as they expose the great work of the Windows Stress team to the outside world for the first time.

Listen: Interview or Download (.wmv)
News source: Channel 9

Report a problem with article
Previous Story

Cancelling WGA Installation sends a report to Microsoft

Next Story

Microsoft looks for better way to search the Net

12 Comments

Commenting is disabled on this article.

Aero Ultimate said,
The real stress test is still coming (when you install Vista on your PC) - for you, not for Vista x)

Most stress-free installation I've ever had on a Windows box.

Come On - Windows 95 was 12 years ago and Windows Me was almost 7 years ago. The whole IT landscape was different then - there weren't as many of the security concerns we have now, everything wasn't connected to everything else in terms of networking and the internet the way it is now.

Anyways, back to the topic in hand - building Windows on a daily basis and what is involved.

Guys - the article above on Channel 9 is all true. Stress runs every night on a large scale (appropriate to the product) for every Microsoft product currently in development. For Windows certainly a build is generally compiled every single night with a sequential increase in the build number. And it's not just the main build lab that do this - it's all of the Virtual Build Labs too (Networking/UI/Core etc etc etc).

VBL's are one level away from the MAIN build in the source code tree structure - it has to be this way or too many devs check in code changes to the main build --- which potentially break each other -------- the MAIN build then encounters build breaks often and CHAOS ensues as people can't work (there are around 7500 in the Windows Division between Core, Client and Server).

Back to VBL's ---- When their code check-ins are approved by the Release Manager for that VBL, they are checked in and a VBL specific version of their Windows build containing their latest changes is compiled and they test run any necessary tests and stress for that VBL is run.

If all is good with the new code check-in's changes made to the VBL builds - the VBL Release Manager goes to the Windows Release Management Meeting to ask to have their code changes from their VBL Reverse Integrated into the MAIN build/source code tree. The release manager also takes the stress repports from the previous night run on the last MAIN build.

Of course --All dev and test--- (and not just the VBL in this scenario) throughout the Windows Division (Core, Client and Server) will then run stress on this build from the latest MAIN build and report the results at the Windows Release Management (WAR )meeting which happens every morning at 9:30am. Each team brings their code check ins to be approved by the release manager in this room (All being well - the VBL Release Manager will get permission from the MAIN Release Manager ****responsible for delivering the overall product to market "when the quality is right and its ready"****.

Once the Release Management meeting is over and all accepted check-ins, the daily build for that day is compiled and is usually available for stress to begin at anywhere between 3-5pm the same day

If someone checks something in that breaks the MAIN build (rare these days in the main build because breaks are usually found during the VBL builds of Windows ---- before it ever gets to main). But if you're a dev that does break the MAIN build - woe betide you - and good luck

From a scheduling perspective, there are milestones (some internal/some external) and the closer the team gets to a milestone, the higher the bar gets for check-ins --- for an external build - CTP, but especially a late beta or Release Candidate, the bar gets very high and you'd better have excellent justification for the checkin. This is the way it has to be to get a quality product out the door. And IMHO Windows XP (especially with SP2 on), Windows Server 2003 and Windows Vista are solid products. There will always be software defects (50+million lines of code in Vista allegedly - so come on!!!!!!!)

My machine is a MAC currently, but plan to move to two machines - a Vista Media Center DVR and a Vista Toshiba Tablet.

Peace


John

Neobond said,
Thats why they never found the 49.5 day Windows ME reboot (BSOD) issue, new builds, patches etc etc..

And the 30 days windows 95 issue...

Thousands of machines spin up an insane stress test that sucks up all available memory and handles..

sounds like normal vista booting