Is there a way to get a rough estimate on how many users you can have taking a survey at once on a given machine? If not is there a way to horizontally scale as needed if you get a large demand?

There are so many factors that play into this that it is basically impossible to answer. Factors that can influence this:
- Type of server (dedicated, vps, shared, etc.)
- Type of virtualization
- Ram
- Processor
- HDD vs. SSD
- Network connection of the server within the data center
- Network connection of the data center to the back bone
- Neighborhood of your server
- Number of surveys active on your server
- Length of surveys
- Question types, Expressions, Quota, etc.
- etc.

Maybe if you give some info about your setup and the expected number of concurrent surveys and someone might be able to respond with practical examples. Also have a look in this forum, because there have been similar discussions before. But remember, with that many factors that can differn in each setup, this is no exact science.

Unfortunately I am not able to get as much info on the hardware as I thought. I'm hosted by GoDaddy and they won't give me any status on the CPU.

- Type of server (dedicated, vps, shared, etc.)
VPS
- Type of virtualization
Not Sure but I think the CPU power is shared (This alone can make everything extremely hard to calculate.)
- Ram
4GB
- Processor
GoDaddy Does not provide this info
- HDD vs. SSD
HDD
- Network connection of the server within the data center
Don't Know
- Network connection of the data center to the back bone
Don't Know
- Neighborhood of your server
Don't Know
- Number of surveys active on your server
No Idea
- Length of surveys
500 Questions
- Question types, Expressions, Quota, etc.
Only 1 quota for 0 for the first question
A good bit of custom JavaScript which should be client side processing.
About 500 questions with roughly the following distribution
List Radio 9%
Short Free Text 27%
List (Dropdown) 31%
Multiple short text 3%
Multiple choice 1%
Text Display 12%
Dummy Questions (not for answering)(for scipts) 15%
A lot of validation equations and some long group skip patterns such as
"((Cargo5.NAOK == "N")) && ((TOKEN:ATTRIBUTE_1 >= "5") and (nextTruck.NAOK != "2") and (nextTruck2.NAOK != "2") and (nextTruck3.NAOK != "2") and (nextTruck4.NAOK != "2") and (HaveTruckAlt2.NAOK != "N") and (HaveTruckAlt3.NAOK != "N") and (HaveTruckAlt4.NAOK != "N") and (HaveTruckAlt5.NAOK != "N"))"
Only 1 hidden quesion thats based of a simple expression (grabs the language user is using)
- etc.

My bet: you will get 0 (zero) completes...

But now serious: I don't know the exact structure and how many questions have to be answered on average per respondent, but 500 questions? Even if on average only 10% need to be answered, who will do this?

The memory of your server is relatively solid, but I still wouldn't send too many respondents at once to the server. I don't think that Godaddy is the very best choice when it comes to Limesurvey (I remember a few problems a while ago, might be solved now) and they are usually also not he highest quality provider.

Do a soft launch. Send out a few invites, check the response rate, see the impact on hour server. Then you might be able to extrapolate from this expierence. Send a few more and check, and so on. Don't just send out all invites and wait.

Also how you will invite people, your target group and the topic of your survey will have an impact. How many people will you invite and how?

My bet: you will get 0 (zero) completes...

But now serious: I don't know they structure and how many questions have to be answered on average, but 500 questions? Even if on average only 10% need to be answered, who will do this?
If I had to guess on average its 15% answered but there are incentives involved and we already have roughly 1500 surveys taken.

The memory of your server is relatively solid, but I still wouldn't send too many respondents at once to the server. I don't think that Godaddy is the very best choice when it comes to Limesurvey (I remember a few problems a while ago, might be solved now) and they are usually also not he highest quality provider.
What type of hosting would you suggest?

Do a soft launch. Send out a few invites, check the response rate, see the impact on hour server. Then you might be able to extrapolate from this expierence. Send a few more and check, and so on. Don't just send out all invites and wait.
We already sent out 5000 snail main invites but our returns are expectant to be lower for postage invites. We still have many many more invites to send.

Also how you will invite people, your target group and the topic of your survey will have an impact. How many people will you invite and how?
Thats hard to say lossy 90,000 but really its closer to 10,000 if your considering more solid invites over the course of 2 years. We are roughly 25% done. Recruiting process usually consists of phone calls then an email invite but we just started postage mail. Recently we have had 3-6 completions a day. This should continue solid while we add the additional paper email respondents.

15% response rate is very good. OK, with good incentives things are changing, but 500 questions... still can't believe it.

> What type of hosting would you suggest?

Well, for such a big and important project I would go with a more professional hosting company. But this is always relative.

3-6 completions a day is nothing for a VPS, you should be able to have easily 10-20 concurrent responses without any major problems. But with 3-6 completions per day, how did you reach 1500 interviews???

Oh actually I meant the average person who takes the survey only hits 15% of the survey. I'm not sure what our response rate.

> 3-6 completions a day is nothing for a VPS, you should be able to have easily 10-20 concurrent responses without any major problems. But with 3-6 completions per day, how did you reach 1500 interviews??

We had more people calling at one point. Right now we are split between several projects so the amount of surveys are a lot slower.

Thanks so much for your guess I really do appreciate it.

Oh, ok. 15% of 500 is 75 questions. Still a lot, but with incentive and if the topic is relevant to them, they might make their way through it. Just don't rely on the data from question 25 upwards too much...