8 Replies Latest reply: Jun 29, 2010 4:19 AM by Lars Helmer RSS

    QlikView Server Installation on multiple servers - possible?

      Hi All,

      We're having performance issues after upgrade to QV 9.0 SR3; it seems to me kinda strange for a machine with 8GB RAM, a descent processor (CPUs are impossible to neither upgrade nor add) and for an online file of about 35 MB, qvs process is taking over 1GB of RAM;

      My questions here are two:

      - is the size of the file really greater than needed for a normal on-line viewing? I know that depends on the data structure inside, etc, but still - what is the biggest file you've launched online?

      - is it possible that the QlikView server is installed on several servers? Here, I mean only the server - I read some stories about splitting the server and the publisher (though the latter is part of the server in latest versions) on different machines, but my questions here is: may QlikView server itself be split on two PCs, so the processor load is somehow distributed between them? If so, I would be glad to learn more - no matter whether use cases or some specific documentation on that.

      Anyway, any recommendations on these performance issues would be great - even those excluding HW upgrades/architecture redesigns.

      Kind regards,

      Peter

        • QlikView Server Installation on multiple servers - possible?
          Lars Helmer

          they way i understand it you want a single document split between two qvses running on different servers (for example, qvs1 loads 40% of the document and qvs2 loads 60%). this is not possible, a qvs always loads and serves the complete document.

          cheers,

          lars

          • QlikView Server Installation on multiple servers - possible?

            Hi Peter,

            basically it is possible to use a QVS-Cluster with QV9. It means that you have two and more server with a running QVS on it. In the management console you can cluster them. After that it can be used as it would by one QVS. You dont have to decide on which server the file should be opened.

            At the moment we use four server; one which controls the management console and web server for QV and three with QDS and QVS (even if only the QVS is clustered).

            I hope you can use this information for you further thoughts.

            Greetings

            Peter

            • QlikView Server Installation on multiple servers - possible?

              I would suggest that we put the idea of clustering and running different components of the QlikView platform upon different servers aside for one moment and address the issue of memory usage on one server.

              You should not need anymore memory when hosting the document for "normal viewing". If you mean the difference from running a qvw on your desktop QV Developer to hosting it within QVS, you should find the RAM requirements to be similar. You need to allow for about 5% of the total RAM used by the QVW per extra user.

              I have an idle QVS and it is using 3.5GB of RAM for the QVS service. Might seem shocking as idle = no users, but QVS is holding a fair QVWs in RAM.

              If possible shutdown the QVS service and restart it, and see how much memory you are using. Also inspect the document settings for "preload" option, since you may be preloading a lot of QVWs and this is affecting the total amount of RAM used for your 35mb qvw.

                • QlikView Server Installation on multiple servers - possible?

                  Hi and thanks for your reply, I am currently researching also in the cluster idea.

                  Now, to your point - after restart, qvs uses twice less RAM, which is quite interesting, however seems normal, having in mind the number of people using it. As for 'preload' option, it is disabled for all documents on the server - if it is settable anywhere in the document settings, please let me know.

                  Otherwise, the same file on my local PC consumes three timess less RAM, than running under QVS. Of course, the server runs also other documents (needed for reloading), however this does not in any way explain why the memory usage doubles a few hours after qvs is started.

                  Guess next steps here should be clustering - if anyone has 'best practices' or similar, or any ideas for performance improvement, please feel welcome to comment.

                  Thanks for your time,

                  Peter