Qlik Community

Qlik Support Techspert Thursdays

Here you'll find the recordings of our Qlik Support Techspert Thursdays sessions.

Highlighted
Not applicable

Troubleshooting performance issues in QlikView Server

Scaling, clustering and recommendations on hardware suited for QlikView Server

For best viewing, we recommend you set the highest video resolution.


Q&A -Troubleshooting performance issues in QlikView Server

Q: ­does the size of .meta file impact the response time of the Access Point ?­

A: Yes, a very large .meta file might impact loading times. In DMS mode the .meta contains the user access list and wil have to be scanned to determine which users to present the document to.

Q: ­Which working limits of RAM are best recommended?­

A: The default Working Set limits (70% / 90%) is a good place to start. Adjustments can be made primarily if you have a server with lots of RAM, as the reservation of 10% for other processes might leave lots of RAM unused when it can be better used to optimise end user experience by containing cached result sets.

Q: ­Can you please advise us where we can find a list of event log error messages and their meanings? Sometimes they are cryptic.­

A: The warnings related to performance are fairly self-explanatory, since they normally say that you are at or above the working set limits.

Q: ­"The Server uses too much CPU" ... if you need to start looking into audit logs to find out what is using the resources at a specific time, do you need to have extensive logging settings / high event log verbosity enabled?­

A: When troubleshooting any issue in QlikView, you want as much information as possible. Setting the logging level to high is the first thing we ask you to do, and also to split the logs daily. For the Audit logging, we generally do NOT need “Extensive Audit logging” enabled, as that will just show all values in a selection. Normally we are satisfied how many of them were selected.

Q: ­And also, how does the size of the SHARED file affect performance?­

A: The size of the .SHARED file will directly impact the loading time of the QlikView document into memory, as it is loaded at the same time as the document.

Q: ­Virtual machine vs physical server - any difference in performance?­

A: Virtualisation always comes at a cost. By default virtual servers share resources with other guests on the same host, and that leads to QlikView having to negotiate for resources as usage goes up. That is a bottleneck. By making sure the server running QlikView has dedicated resource pools, most of these limitations can be overcome and you can come fairly close to the performance of a dedicated physical server. Setting the correct expectations based on the choice of Virtual/physical hardware is key.

Q: ­Can i restrict the number of cpu used for reload documents­

A: While not a part of this session, the topic might have impact on QVS performance. By default a reload engine will consume as much resources as it can get, and might thus impact the QlikView Server and the end user experience. For this reason we recommend setting QlikView Publisher up on a separate machine, to allow both roles to be as efficient as possible, ensuring both maximum reload performance and maximum end user experience.

Q: ­Maybe a quick word about the different options of load balancing?

A: Great question, should have gone into a bit more depth. The AccessPoint can use 3 different Load balancing algorithms in the connection to a clustered QVS. The default, “CPU with RAM overload” will simply send the user to the least loaded QVS server, but will prefer one with the document already loaded into memory. The second one, “Loaded Document” will send the user to the QVS where the document is already loaded into memory. This setting in combination with selecting which nodes in a cluster to preload individual documents on gives great granularity in load balancing manually. The third option, “Random” is just random. J

Q: ­In QMC, how does the reload document work? is it affected by RAM or CPU?­

A: Running a reload spawns a reload engine process, a QVB.exe that performs the reload. That process will, as any other, consume resources. Depending on the reload script and how much aggregation is done on the data retrieved, the actual load on CPU and RAM might differ. The “Task Performance Summary” introduced in QlikView 11.20 SR7 gives great insight into the resource consumption on task level. How to set it up is cover in the release notes from SR7 onwards.

Q: ­How much does disabling NUMA in the bios increase performance, vs using the softswitch in the INI config file to disable it?­

A: From the introduction of the automatic NUMA soft switch in QlikView 11.20 SR2, where the QVS can detect the NUMA environment it exists in, we see very little difference in real performance between the soft switch vs. disabling NUMA in BIOS.

Q: ­With QlikView Enterprise server without Publisher, what are the settings of set work min and max recommended?

In other words, QlikView Server and the scheduler in the same server!­

A: Without Publisher, the recommendation to use the default Working Set limits stand. We would rather see that reloads are taking a bit longer and that end user experience and performance is prioritised.

Q: ­Could any memory/cpu insufficient issue cause a reload task failure? (The qvw involved in the is reloaded and saved successfully but error that triggered the failure is "Could not close the SourceDocument.")­

A: This issue should be raised as a Support ticket to be fully investigated. In general terms yes, the error message could very well be a result of resource starvation.

Q: ­Should QlikView server automatically add that flag into the ini file?  i had a recent case where qlik told us to add the flag explicitly.  We had not added the flag because we were under the understanding that you just mentioned ...

A: Adding the NUMA switch manually to the Settings.ini file was only needed prior to the introduction of QlikView 11.20 SR2. Any environment where this has been done, adding the soft switch manually in the setttings.ini should be revisited after upgrading past SR2 and the line should be removed, allowing the QVS to intelligently adapt to the presence or absence of NUMA.

Q: ­We have recently upgraded a QlikView installation from 11.2 SR4 to SR10, afterwards QMC is not functioning well: very slowly, users in USA are not able to log on QMC and get the message "Processing...". The servers are located in Denmark.­

‑­­Other part of system are functioning without any problem (Accesspoint, running scheduled tasks...). I contacted Qlik Tech and they recommended a reinstallation of the new version SR 10, but it has not helped the issue. ­

A: This issue should be reported to Qlik Support as a separate ticket and be investigated accordingly.

2 Replies
Not applicable

Re: Troubleshooting performance issues in QlikView Server

solid gold ... for some of us

Re: Troubleshooting performance issues in QlikView Server

Thank you!