DNS performance benchmarking update

on December 14, 2009 in Linux with no comments by

A few days ago I posted a short blog about DNS performance benchmarking by Google using namebench. After the rather surprising results, I decided to replace the primary upstream DNS server with OpenDNS.

Once I had run the namebench too again, the results were showing a huge leap in performance. But strikingly, BT’s DNS servers were still faster than my own local servers. Jonathan, the resident monkey, also commented about this based on his own tests. It seems that there might be something amiss with the namebench tool (and I shall dutifully point this out to Google, to see what they have to say).

Nevertheless, I had run the namebench several times and it did indeed radically change from two days ago where my own service provider’s DNS servers were the upstream. Following is a chart from the last benchmark:

Namebench Chart (Updated)

Namebench Chart (Updated)

As you can see, my primary (SYS-10.10.24.4) and secondary (SYS-10.10.31.1) are now right at the top of the chart, showing a slight slow start for the primary. Unfortunately, my service provider’s DNS server (in orange) was once again trailing everyone else in this test. In fact, it did so in all the tests.

Following is a chart showing the mean response times in milliseconds:

Mean Response Times (Updated)

Mean Response Times (Updated)

And this (chart) is the odd part. You would expect that a server on the same network, solely used for DNS and with less than 5% average usage, would perform better than any external server once the DNS responses have been cached.

While the primary DNS server is now faster than the poor performing DNS server of my service provider, it is still lagging behind BT and even Google’s public DNS servers. I am curious to know about its cause!

In all, namedbench reported that my primary server is 89.7% slower than “BT-30 GB”. Up from the 274.4% a few days ago – an improvement indeed!

Join the discussion

Your email address will not be published. Required fields are marked *