Recently we where asked what spec hardware should we buy for our new Data Collection Interview Server setup, and we had to honestly reply we don’t know, but we know people that might have an in stite on some rules that they used. So … When you purchased your setup , how did you decide what hardware to buy? Did you use any sort of sizing calculator or was it just a guess?
If you did not use any sort of calculator, but have a setup in use and its working for you, what spec machines do you have , how many emails do you send out , and what is your response rate for WEB and CATI ? How many questions do you have on average in your surveys? What does DimensionNet or Interview Administrator say your max con-currents are ?
We would love to hear from you be it from a comment on this article or a email to admin@smarterDimensions.com if you are willing to send us all your IVW* log files ( from all servers in the cluster ) then that would be great , we plan to run some articles on how we can write programs to read these and work out actual concurrency and then perhaps come up with a sizing calculator. So anything you send us will help us write a realy cool set of articles and scripts. We promise that all confidentiality will be kept and no names will be given as to whose setup is who’s
Regards
The Smarter Dimensions Team.
PS We cannot do this without your input !!
When I talk to customers about sizing , the first thing I work out is the required simultaneous connections, this is done roughly using the following calculation. This is based , on a perfect world and even distribution, but it does begin to give you insight as to what may be required.
Let’s say 100,000 emails sent each year with a 30% response rate.
So that 30,000 started interviews per year.
Which is 82 starts per day.
Which is 4.8 per hour in a ( 17hr day )
So if the survey has 1hr long , that’s only 4 simultaneous,
There are lots of other things to take into account, but it gives you a heads up and helps me make a call.
When it comes to hardware , I normally say how much money do you have , and how redundant do you want the system. There is allot to think about here, but simply put for IIS I recommend 4 gig ( 3 session engines ), at least 80-100gig hard disk, SQL 8gig , at least 200-500gig. As for architecture, that really does depend on money and required simultaneous connections. The more hardware you have the better to be honest.
Hope it helps and look forward to the articles around this topic.
We have used a one machine solution mrInterview/SQL combination up to now. Our interviews are 30 or so minutes long.
What we have found is that when our 4 core and 4 gig memory machine hit about 2500 interviews a month the build process on the machine crawled to a stop. We would commonly have 3 interviews being built at any one time.
We upgraded to 2 quad cores with 4 Gig of memory. This does well from a standpoint of handling interviews, but we instituted a live results script using mrTables which recalculates the sum of each interview’s responses after each interview and publishes them. This has bumped up our processor utilization to 30 – 40% fairly regularly. The build process is the part users notice. The interviews seem to run rapidly even under that effort.
We currently peak at about 5000 interviews a month. Our next solution will most likely be a separate maching for the Sql Server part.