Comparing the performance of AppFabric against Sql Server

Comparing the performance of AppFabric against Sql Server

I’ve been doing a lot of work lately implementing distributed caching systems for various clients. During my initial scoping, I found a lot of information out there comparing the performance between cache types (AppFabric, Memcached etc), however I could find very little comparing the performance of caching vs the actual database (in my case, AppFabric vs Sql Server 2005), just that it’s “much better”.

Despite this lack of statistical information, I went ahead with caching anyway (after all, ‘much better’ sounds pretty good), and because I’m a Microsoft shop, I left it tidy by selecting AppFabric to ease the load on Sql Server.

Now that I’ve implemented the code to a decent extent through a particular website, I’ve been able to conduct my own performance benchmarks and here I’d like to share the results.

Note in particular that these are what I would call real-world results. I didn’t attempt to isolate cache access based on particular SQL queries or size of the item. I didn’t reset the cache between web pages. I simply used the website in the same way I’d expect my users to, and recorded the overall times.


My methodology was fairly simple and certainly prone to a margin of error:

  • installed a ASP.Net website to a development machine which I knew I had sole access to. Sql Server is installed on the same box as the web server, however I had a separate (and dedicated) server containg the cache.
  • switch the cache off (via AppSettings configuration)
  • browse through a set number of web pages, without refreshing the client browser or anything tricky
  • made a note of the pages I travelled through, then turned the cache on and went through the same pages again

I had simple code around my ‘cache access’ block which simply counted the number of milliseconds the application spent trying to Save/Load the items (or bypassing if the cache was off) and logged them to a text file.


Data access is split into two parts – Save and Load. Note that Sql Server does not have the burden of a Save method, and AppFabric of course only calls Save the first time the item is accessed.

Interesting results:

      • AppFabric increases my data-access speeds by almost 5x over Sql Server. Thank goodness for that, and of course the longer you cache an item, the greater this efficiency will get. Again, this is not saying that AppFabric accesses data 5x faster than Sql Server on a call-by-call basis, it is the overall time benefit of implementing caching
      • AppFabric has an additional overhead in saving a newly calculated item back to the cache. In my case it was 2,265ms for the 81,960ms I spent loading items – a ratio of about 1/36. The longer I ran the test (or cached the items before they expired), the better this ratio would have become

To conclude

AppFabric clearly has performance gains over Sql Server – in my limited test it made my data access 5x faster, and a longer test expiry period would have made this much much greater.

Note also that the website I was using had (test) database tables less than ~1,000,000 records, although some fairly funky SQL queries are being made here and there. For an even larger database such as Facebook (or my client’s database, hopefully) the SQL queries would take longer, but (I suspect) the caching times would remain exactly the same – another point in favour of AppFabric.

Uh-oh…one more thing (and it sucks)

Now that you’ve read this far, I’ll tell you the real reason I ran these tests, which is that I was convinced that AppFabric actually made my site slower. Not because SQL was better, but because the previous caching system I switched out was in fact the good old ASP.Net HttpCache utility. Like AppFabric, this cache is entirely in-memory but because it doesn’t concern itself with regions, tags, (and much more), it actually runs much much much faster. Let me type that again so Google picks it up – AppFabric is not nearly as fast as the built-in ASP.Net HttpCache utility. I ran the same test as above, but using the HttpCache:

Yes, that’s right – AppFabric is 2.1x slower than the built-in ASP.Net caching utility. Very sad, especially when I see my site slow down after all my hard work. However, I’m sticking with AppFabric:

  • HttpCache resets whenever the IIS App Pool resets, when you redeploy and any other time it feels like it. I suspect if I ran this test over a long time (say, a week) then the results would be closer
  • There is no tagging in HttpCache – you have to roll your own by integrating into the key – and the subsequent parsing to ‘find by tag’ is slow
  • HttpCache cannot be expanded easily to hold gigabytes of data – it simply won’t work when my sites expand
  • The distributed nature of AppFabric allowed me to build a separate ‘admin’ tool where I can look into the cache from another website, count it, clear it etc
  • And if I’m honest, the final reason is that this type of caching is all in vogue at the moment and it’s something I feel I should be part of

One thing I’m wondering about is actually using a combination of both – extremely high-access queries (like user permissions) could be HttpCache, leaving AppFabric to handle the larger datasets. Bit of an art form I reckon.

Automatic code documentation based on your C# comments

Automatic code documentation based on your C# comments

Best practice architecture for professional Microsoft.Net websites

Best practice architecture for professional Microsoft.Net websites