Forum Discussion

pierrevocat_109's avatar
Icon for Nimbostratus rankNimbostratus
Oct 30, 2008

DNS_Request redirects traffic slowly




I'm using the following IRule (V9.3.1) to direct traffic between 2 physical sites/locations (A and B) and also redirecting traffic based on whether it is internal or external.





if { [IP::addr [IP::remote_addr]/16 equals]}{


if { [active_members gtm-pool-siteA-internal] } {


pool gtm-pool-gtm-pool-siteA-internal}


else {


pool gtm-pool-siteB-internal}


} elseif { [active_members gtm-pool-siteB-external] } {


pool gtm-pool-siteB-external


} else {


pool gtm-pool-siteA-external







I've set up 4 GTM pools and it seems to be working however when the pools are marked as down it takes clients a while before they get redirected to an alternative pool instead they keep on hitting the pool that is marked as down until I flush the clients dns or the browser is closed and opened up again.



Is there any other alternative to DNS_REQUEST.






5 Replies

  • What is the TTL set for on the pools? Most LDNS servers out there should honor the TTL, as do most operating systems. If you are using Internet Explorer 4,5, or 7, the DNS records are cached for 30 minutes regardless of what the TTL is, so the browsers will need to be closed and reopened if the stale record is present. This can be changed in the registry, but that's not always practical. IE6 only caches CNAMEs, so if you are using A records, it uses the system cache, so your TTL will be honored. Firefox 1.5 and later cache for 1 minute, so the recovery is much faster.



    On your question, the event you want for GTM is the one you are using.
  • Thanks citizen_elah,



    The TTL on the pools is set to 30 seconds, unfortunately most of our customers use IE6 and IE7.



    Do you know if the GTM's Topology settings would work in the same way i.e. the page not being redirected for however long the DNS records are cached for?



    Thanks again for your help.




  • You can confirm that the GTM is making the switch immediately by adding log entries to your rule and making a request from a new browser window on a machine that isn't caching at all. Unfortunately, the failover time is dependent on the LDNS infrastructure of your clients and their choice of browser, so topology won't speed things up.
  • No problem. There's always a tradeoff when planning for failure. How badly do you want to impact the customer experience to plan for the occasional outage? By keeping TTL's really low, particularly in a delegated scenario, each request must be made twice (once to the authority, then to the GTM) and in high latency environments this can impact the perception of the overall performance. A higher TTL may result in a longer outage, but won't impact the day-to-day of normal operations. Knowing your audience and the landscape in which they reach you will help in finding that sweetspot.