Hello everyone,

i have some performance (?) issues with my limesurvey installation (Version 2.56.1+161118).


I am using 2 surveys in a row (with a different external website in between).For at least 5% of the participants either the first or the second survey does not load at all. If they load, they usually take 10+ seconds to load the first page (later pages are 1-2sec).

Other sites load very well (quick and reliable) on my server, so I'm not sure if it is a webserver or limesurve issue.

I will try a different webserver later today, but I was wondering, if this is a known issue with LS (couldn't find any existing threads, sorry if I missed something).

I thought about using limeservice, but I need ~1000 participants (so ~2000 surveys), so the provided packages don't really fit my needs.

Any other ideas, how to increase the reliability and performance?

Best regards and thank you
Andreas

LimeSurvey is creating a big session file during the first access. Can slow things down when you have a lot of users and sessions files are e.g. created in /tmp.
Check the session directory of php.ini and set it to a different directory if it set to default (tmp). Slow storage (I/O) is adding up when it comes to session handling with LimeSurvey.

When bridging surveys you always loose a few users via the redirection.
Do you loose the 5-30% (how did you got those numbers?) on accessing the first Limesurvey (LS1) or access the website (Web) or accessing the second Limesurvey (LS2).

What kind of OS, Webserver, PHP and database version is used?
How many users are hitting the server per hour, per minute?
How did you announced the survey URL? TokenLinks? PublicLinks (E-Mail or on websites?).

Thanks for the session folder information, I'll check that. It could very well be a storage/session problem. As my own web site writes to the (same) db, but does not use sessions, this might explain some of the difference in speed and reliability.
I will try to change the session folder and perhaps use different sql databases for the surveys and the other website.

The 5%:
There is some loss on accessing LS1 (can't track that reliably), no loss when accessing my Web site and an average drop of ~5% from Web to LS2 (data from LS1 and Web site are logged, but LS2 session is not initiated). I log the redirection Web->LS2, so the LS2-link is definitely called - it just times out (or takes very long to load, so some users might actively drop out).
The number (5%) went up in my most recent test on mTurk - this might be due to parallel load. Lost up to 30% from Web to LS2 and had several complaints of LS1 not loading. But load was below 20 users at any time.

Load:
Currently not more than 20 users at the same time. Probably max 5-10 people accessing LS1 or LS2 at any point in time. So 5-20 per Minute. Not more than 30 per hour.
Ideal target would be ~200/hr, perhaps 20-50 at the same time. I need a total of around 1000 users over less than 3 days.

survey URL:
Public link to LS1 via mTurk (and other channels).
LS2 is only accessable via (public link) my website and not published elsewhere.

Server Info:
it's a hosted/shared server (strato), I don't have access to exact performance data. German server also for US participants - this might also be an issue.
OS: some *nix - don't know for sure
php: 5.6 (server also supports 7 / 7.1)
mySQL: 5.6.35

I'll try to move LS to a different server + DB tonight and see if it works better.

You also don't run the most recent version of LS, so if there are some performance problems due to a bug they might be solved in a newer version.

If you redirect people, you will always loose some of them, not only due to technical issues, but whenever there is a break in the survey, there is a good chance that people will just drop out.

There is no visible break between the parts - it's made to look like one large study. The only "break" is the time it takes to load LS2. And even if people drop there, it should at least be logged in the LS2-data. Almost all dropped participants are redirected to LS2 (I log the redirection web->LS2) but never "arrive" at LS2 / no LS2-session is established (there is no data logged for LS2, not even a startdate).

I'm rather certain that it is a technical issue (or at least the main part of it), as I had troubles in my own pre-tests aswell, just not as severe.

Sometimes only 1 user is enough - and LS1 or LS2 doesn't load. So it can't be only the server/session load (only if my hosted server is really that bad - but never had issues with other applications). It just gets worse with more parallel accesses.

I'll try to update to the current version, perhaps this also helps - thank you for this hint I tried to avoid the upgrade, as I made many changes to the css-templates.

Strato shared hosting means that DB is on a different server that the webshost?
I/O performance on Strato is often "suboptimal".
Everything is crammed into the session file. So a file is often 1MB or bigger. That is created when people hit the server. Check difference between a 1 question survey and your real surveys. Any difference in loading time till you get to the first question?

I am not certain, but I think the DB is not (necessarily) on the same server as the webhost. The DB performance seems to be fine. I'm currently trying to move LS to a different server with better I/O.

Smaller surveys load a bit quicker (1-2 sec). Right now everything is fine ... for me the LS2 survey loads fairly quick aswell (2-3 sec). Just sometimes the time strongly increases or it doesn't load at all (for some people). Currently for mTurk over half of the calls seem to fail, which is really weird. Might be an DE/US problem.

This is a tricky situation and I think it will not be easy to identify the problem, because at the moment we have so many possible variables and so far I think we could not rule out many of them as the cause for these problems.
- Limesurvey bug: Possible, but not too likely as the majority gets through, I haven't seen similar complaints on the forum.
- A local server issue: possible, because with shared hosting you never really know what you are getting and who your neighbors are. It could be that some other user is bringing the server down or that your load is actually too much for the server/package configuration at Strato.
- Eventual problems between US/DE: also possible. I am located on the southern hemisphere and sometimes we have problems or slowness to access servers in Europe or the US, while other services that use servers distributed worldwide, don't have these problems. But this should usually temporary and between Germany and the US there should be relatively good connections. So if this happens regularly, I think the problem lies elsewhere.

How about your two surveys? How are they structured? A lot of questions? A lot of equations or conditions? Anything special, custom made, different from the usual?

ahog wrote: Currently for mTurk over half of the calls seem to fail, which is really weird. Might be an DE/US problem.

Might be a mTurk audience problem as well.

LS1 is very simple
- welcome page
- 1 group with 1 text question + one hidden equation (calc. treatment group id using userID).
LS2 longer:
- 7 question groups (on 7 pages), ~120 questions (incl. subquestions) - mostly array and some text fields.
- end page.
- 3 hidden equations (one with 11 IFs and 8 RAND - the other two much simpler). The equations build on each other and the last equation is referred to in a text.
- no conditions, nothing custom or other special I can think of.

I'm contemplating to remove LS1 and build its functionality on my own platform - and host LS2 on limesurvey pro. This should be quick enough and remove potential problem sources. Just a bit expensive and inflexible for only 1000 responses.

Perhaps moving to a different server will solve the problem, will know tomorrow.

In any case, thank you both very much for your support. Much appreciated!

(might also have - in addition - to do with the mTurk crowd ... but I've had the problem before with other participants and on my own machines, so there is definitely something wrong on a technical level - and dropping doesn't really make sense at this specific point in the study flow.)

You can purchase 2x the basic package with 1000 answers, that shouldn't be too expensive. Should be around 50-60 Euros or something (the prices I see are in a different currency).

That would work, I'll check it out later.

Are there any US hoster that are know to run LS well?