Scott Hanselman

Greg Hughes and I stayed up late last night tuning and installing a custom build of dasBlog. If you remember, dasBlog is Clemens' rewrite/imagining of Chris Anderson's BlogX codebase that has been moved over to GotDotNet and is under the supervision of Omar Shahine.

ORCSWeb, my awesome ISP, and Scott Forsyth (ASP.NET MVP who works there) had complained to me that as my traffic increased, my website instance was being a poor citizen on the shared server. My site is on a server with something like a dozen other sites. While I'd survived slashdotting, my traffic lately has been getting big enough to bother the server.

ScottF had noticed that my blog had these unfortunate characteristics (remember these are bad):

• CPU Threads that were taking minutes to complete their work
• Caused 9,538,000 disk reads during a time period while another site on the same server with twice as many visitors had 47,000 reads.
• My process was tied for CPU time with "system."
• I used 2 hours, 20 seconds of CPU time one day. My nearest competitor had used only 20 seconds.
• I was 2nd for disk reads, and 11th for disk writes (the writes weren't bad)
• In a day, I surpassed even the backup process which was running for a WEEK.

These bullets, of course, are quite BAD. So, during my recent burst of creativity when I added a number of features to dasBlog including a comment spam solution, a referral spam solution, and an IP address blacklist, I did some formal performance work.

If you're familiar with dasBlog, I yanked the need for entryCache.xml, categoryCache.xml and blogData.xml, which were older BlogX hold overs, and move them into thread-safe in memory storage. I change the EntryIDCache and other internal caches, and added outputcaching for RSS, Atom, and Permalinks.

According to ScottF and the folks at Orcsweb and their initial measurements "from what I can tell today, this *is* 250% better. CPU used only 20 minutes [as opposed to nearly 2.5 hours] of time by the end of the day and disk IO was much less than normal." This is early, but we'll see if these numbers hold.

I seem to have a few other bugs to work out, so hollar at me if the site's goofed, but otherwise I hope to get Omar to integrate these changes into his own great new stuff coming in dasBlog 1.7.

During this perf test, I used Perfmon, CLR Profiler and other tools, but mostly I thought. Just literally say down and thought about it. I tried to understand the full call stack of a single request. Once you really know what's going on, and can visualize it, you're in a much better position to profile.

Since you are a technical group, here's a few tidbits I found during this process.

• If some condition can allow you to avoid accesses to expensive resources and bail early, do. For this blog, if an entry isn't found (based on the GUID in the URL) in my cache, now I won't even look in the XML database. Additionally, I'll send a 404, use Response.SupressContent and End the Response.
  

      1 if (WeblogEntryId.Length == 0) //example condition
  
      2 {
      3     Response.StatusCode = 404;
      4     Response.SuppressContent = true;
      5     Response.End();
      6     return null; //save us all the time    
      7 }
  

• Lock things only as long as needed and be smart about threading/locking.
• If you're serving content, caching even for a few seconds or a minute can save you time. Not caching is just wasting time. Certainly if I update a post, I can wait 60 seconds or so before it's seen updated on the site. However, if a post is hit hard, either by slashdot'ing or a DoS attack, caching for a minute will save mucho CPU.
  <%@ OutputCache Duration="60" VaryByParam="*" %>
  at the top of one of my pages will cache the page for a minute using all combinations of URL parameters. To be thorough, but use more memory, one would add VaryByHeader for Accept-Languages and Accept-Encoding, but this is handled in my base page.

About Scott

Scott Hanselman is a former professor, former Chief Architect in finance, now speaker, consultant, father, diabetic, and Microsoft employee. He is a failed stand-up comic, a cornrower, and a book author.

About   Newsletter

Hosting By