NS Upgrade

[Dean]

We have upgraded Bind on all name servers from 8.3.7 to 9.3.

The upgrade has dropped CPU and memory utilization dramatically. CPU has gone from 60-70% to 1-2%. Internal queries are running at >1ms.

This should help reduce execution time a little on scripts that use database calls, mail functions, etc.

[Brangwyn]

Surprised we weren't on 9 already, theres definitely a much smaller footprint with the later versions of Bind ... (some of the early 9. versions were a bit rough though)

Good upgrade

[TheDesigners]

I've noticed a huge speed increase on SQL queries. A site that always had a little lag time because of several queries on each page now is a rockin’! Good job!

[Brangwyn]

I'm surprised you notice any difference really, IP's should be cached by the local DNS Resolver on the servers, so they really shouldn't be trying to resolve internal vortech addresses often enough that it should make any siginificant difference (unless theres more to it that Dean hasn't mentioned .. like some DNS missconfigurations found as well).

[admin]

Yea that would be odd bind would speed that up, it should speed up new visters coming to your site maybe and the time it takes for us to restart bind, but would not think bind would have made a difference in sql speed. That might have came from the network upgrades we did the other night now that 80% of the racks are gige with fiber gige from core switch to routers now.

[TheDesigners]

Without question, there is a noticeable difference. There has always been just a little lag when a page would start to load. That lag is gone now. The site is setup as a CF application and other than a few constant objects (menu, logo, etc.) each page is populated from a MSSQL DB.

Because this company usually has several users updating the DB at the same time, I've had to add: <cfheader name="Expires" value="#Now()#"> into the app, which forces the client browser to purge it’s catch of the requested file, but that shouldn't have any effect on the local server's cache,would it? If the upgrades don’t have an effect, why the speed increase? Is it simply because it’s the weekend and there’s less load on the SQL servers?

[Brangwyn]

This is most likely a result of the new peers more than anything or the box performing better, if you think about it DNS lag can only happen once, on your second page refresh everything including the IP address of the MS SQL Server is already cached, so DNS isn't actually touched again.

Your Expires header will have some effect, it's forcing graphics etc to be reloaded, every page get (though you shouldn't really need an expires tag there imho). I suspect what your seeing is just the server performing a bit better than you'r used too or some improvements becuase of the new peering, perhaps becuase of a lower weekend load, guess that will be proven come Monday though

[Dean]

No misconfigurations, but each web server is doing 25-30 DNS requests per second, some of the bigger servers can do a couple hundred at the peak of the day. If you shave a millisecond or two off of each, then scripts will be executing faster across the board.

The whole point was to show that this could have_some_kind of relevance to resellers.

[Brangwyn]

They'll be executing faster if the load on the box has reduced though, which may indirectly be as a result of the faster DNS timings. An email script for instance executes without any DNS resolving going on, it's the SMTP service once received the email that would then do DNS (if thats quicker, lower load, and potentially server seems more responsive).

Curious though, I was under the impression the webservers don't use the nsx.hsphere.cc nameservers, they would have to be using something else to ensure domains whose DNS is hosted externally are resolved correctly wouldn't they? else you'd have bundles of trouble.