Do ISPs respect DNS ttl?
- Thread starter flysarescary
- Start date
Haha sweet. I want to set a long ass time for ttl hence my questions.
Do you have an estimate/average on what the upperbound of ISP ttl is. What is a reasonably long ttl? 30 days? 60 days?
Do you have an estimate/average on what the upperbound of ISP ttl is. What is a reasonably long ttl? 30 days? 60 days?
I suppose now that I think about it there really isn't one. If ISPs generally maintain DNS caches now that I think about it, what is the point of optimized DNS delivery services. Akamai (I'm assuming they do it), Dynect Platform (considering purchasing). I am looking to route DNS requests pointing users(ISPs in reality?) to the closest server geographically.
My original objective was to maximize the ttl to minimize long dns requests from UK to the US. So the questions that come up are...
Are new DNS records for uncached new domains propogated (requested by the ISPs) to ISPs internationally and automatically? If not (I'm assuming), then the first visitors from each ISPs would have a delay as the ISP/(user?) requests the DNS records correct? If that's the case I would want to minize this with as long as possible ttl.
Are the overwhelming majority of DNS requests made by ISPs?
Also, apart from new domains (which ISPs don't have any records on and require that first request) if ISPs generally maintain DNS caches what is the point of optimized DNS delivery services. Akamai (I'm assuming they do it), Dynect Platform etc if the user still receives the information locally from the ISP. I am looking to route DNS requests pointing users to the closest server geographically.
Do you have any statistics/ratios/estimates between unique clicks/server requests and DNS queries with a TTL of 30 days (or a TTL you have relevant or even estimated data on). Let's say 50,000/clicks a day how many DNS queries would you guys estimate because of DNS caches?
Are new DNS records for uncached new domains propogated (requested by the ISPs) to ISPs internationally and automatically? If not (I'm assuming), then the first visitors from each ISPs would have a delay as the ISP/(user?) requests the DNS records correct? If that's the case I would want to minize this with as long as possible ttl.
Are the overwhelming majority of DNS requests made by ISPs?
Also, apart from new domains (which ISPs don't have any records on and require that first request) if ISPs generally maintain DNS caches what is the point of optimized DNS delivery services. Akamai (I'm assuming they do it), Dynect Platform etc if the user still receives the information locally from the ISP. I am looking to route DNS requests pointing users to the closest server geographically.
Do you have any statistics/ratios/estimates between unique clicks/server requests and DNS queries with a TTL of 30 days (or a TTL you have relevant or even estimated data on). Let's say 50,000/clicks a day how many DNS queries would you guys estimate because of DNS caches?
Long DNS requests from UK to the US? DNS requests are extremely fast. The difference for me to query a US domain vs. a UK domain is anywhere between 100 and 200ms more - a fraction of a second.
If you increase the DNS TTL & something happens with your server, you're fucked if you need to change your DNS records. ISPs could have the wrong DNS entries for days!
DNS queries are fast, are not bandwidth intensive, and it doesn't make sense to have long TTLs.
Also, remember - we're talking about 100 to 200 bytes of data for each request. That'll be the fastest part of your site to load. It's nothing.
If you increase the DNS TTL & something happens with your server, you're fucked if you need to change your DNS records. ISPs could have the wrong DNS entries for days!
DNS queries are fast, are not bandwidth intensive, and it doesn't make sense to have long TTLs.
Also, remember - we're talking about 100 to 200 bytes of data for each request. That'll be the fastest part of your site to load. It's nothing.
Same here. Just moved some sites to a new server, and had some network issues, so I moved the sites back (using a DNS solution external to the server), and the TTL was 3600 for the domains in question. Time Warner and OpenDNS had things updated within 15 minutes, but the 4.2.2.2 DNS server still has the old IP after 20 minutes (oddly, 4.2.2.3 has the new record updated).
...which is why I set the TTL on my domains to 300 for my most recent transfer.
I also find that my ISP (Verizon Fios) updates more frequently than the TTL, as the two DNS servers that they have aren't tied together with a single cache & my router randomly chooses which one to query.
You're correct, they are not propagated automatically. General DNS admin wisdom would suggest having a reasonably long TTL to maintain your hostnames' DNS cache at major ISPs. But not too long as to create a hassle when you need to change the data.
Yes, because the majority of clients (our PCs) query our ISP's local nameservers, which have the responsibility of communicating with the real Internet and resolving the desired hostname.
1. Reliability
2. Reliability
3. Reliability
I've admin'd BIND servers for years and the only reason I haven't switched to a managed solution is because of the cost.
The experts should have a fully redundant setup spanning multiple data centers, whereas I'll have two boxes with redundant hardware at two separate locations.
Also, these guys can route DNS queries to the closest geographic server using ANYCAST routing. Your typical DNS setup simply round robins between the listed name servers. You need to have a registered ASN to be doing this type of routing (typically ISPs or organizations which require ISP-level routing abilities).
There is such a thing as a negative cache, where a NS will cache the fact that it tried to look for an item and received an error or no reply at all. So let's say your DNS server was down for a bit and created a negative cache at a major ISP... even though your TTL might be 30 seconds, that negative cache could potentially stay there much longer.
Theoretically... cache hit rate should be very high 70%+ with a TTL of 30 days.
This is a pretty good skim:
http://nms.lcs.mit.edu/papers/dns-ton2002.pdf
An academic paper that analyzes among other things, the effect of TTL on cache hit rate.