In May this year I discovered DNS Resolution Time problem which was taking around 4 seconds to resolve for GSM tests.
I raised a case with Microsoft who instructed me to enable a VS web test to help diagnose however I managed to get to the bottom of the problem before starting the case.Thought I would blog as this may very well catch me out again.
The Graph are four separate GSM tests from London endpoints for DNS Resolution Counter for the same URL ‘www.blahblah.com’ Over a 24 hour period. Two are displaying high DNS and the other two are displaying less than 1 second so no consistency there which also implies an error with the GSM config.
This graph displays the DNS Resolution Counter over a three day period selecting only the GSM tests in the previous screenshot that were below 1 second. You can see that around 10:00 PM ish UK time the problem had gone away for those two test.
Clearly something wrong here. I ran independent tests with numerous external tools including the obvious human option browsing the site and timing how long the page initially displays. My conclusion was if it is taking 4-6 seconds to wait for DNS as GSM is indicating then it would be incredibly easy to detect and recreate. Needless to say I could not re-create the same times interdependently.
Some companies will use devices for DNS name resolution for client requests. So this problem may or may not manifest itself as a problem for all clients. Alongside onsite network staff have identified the cause is due to GSM making checks over IPV6. The Devices are not configured for IPV6.0 The cause to slow DNS server response times are due to the request from the client over IPV6.0 which eventually results in a timeout.
( Not 100% sure all GSM checks are IPV6 followed by IPV4 but does explain the sporadic results.)
We added values to non IPV 6.0 network device tell the client to respond with NOERROR which will result in the client not checking each DNS server up to root before timing out. Primary cause of the delay
This will lead to no timeouts for IPv6 queries because they will quickly receive the ‘noerror’ response and should therefore generally improve the DNS experience for clients and consequently statistics will also improve immediately. I thought I was really onto something when I said the client have issues with site speeds 🙂
Anyways here is a similar thread which I think is related