Hardware and Performance Optimization
QST was created to scale economically - non geek: runs cheap.
By running on Apache 2.x server using the event MPM module which runs multi-threaded on linux, our mod_perl module QST.pm runs multi-threaded also.
Apache 2.x also runs mutithreaded on Windows 10/11 through winnt MPM and mod_perl runs multithreaded QST.pm on it too. The default configuration of 1,923 threads is production grade.
MySQL runs multithreaded on both linux and windows and the database was designed for maximum performance in conjunction with the software.
Running multi-threaded means significantly less resources and memory are required, and all cpu cores/threads are being used efficiently.
With our Linux performance configuration of: 50 Apache children, each child running 100 threads means you have 5,000 threads waiting to answer requests.
Inside of this we have Apache linked to 50 perl interpreters, each running 50 threads for a total of 2,500 threads waiting for requests.
This requires approx. 365 MB of memory and will grow, but memory is shared/reused and after a child process has handled 10,000 requests it is killed and memory will be freed.
Everything has been optimized for heavy simultaneous use.
All this happens on one machine, no multiple web servers, no load balancers, no cloud, etc. needed.
Upgrading is easy, as all that is usually required is replacing your QST.pm file with the new one.
And best of all, we help you install/tune it if you require (free), and then document it so others can benefit also.
------------------------------------------------
QST will run best on a linux OS.
QST will run on a laptop, server or in the cloud.
- run it on a laptop if you have only a few hundred users and wish it to be portable
and you can plug your laptop into different networks for different groups of users.
We recommend a Core I7 processor with 8 cores/16 threads, 32GB ram and 500GB NVME SSD with a gigabit network card.
*** UPDATE Aug. 5, 2024
We recently updated our QST servers to HP Workstations Z 840's
Each contain 2-Xeon 24 core 48 thread processors, 128 GB ram, 1TB NVME ssd's.
These can be found for bargain prices - Ours were $750.00 CAD each at bestbuy.ca
If you plan on conducting extremely large assessments, these will handle the load.
Only run Apache, MYSQL and QST on the machine.
- MySQL 5.7 will run better because it caches results (MySQL 8 does not do this).
- This only helps if you do not have other users creating questions while the quiz/test
is being run.
Perl likes RAM, so more is always better.
32 GB ram means about 30 GB is available for QST.
Apache and QST's server memory footprint is about 7 MB.
The NVME SSD will ensure the database will have very good response times.
QST is not processor intensive, and all tables in the database have been optimized by indexing.
Lesser configurations will also work, it mainly depends on the number of
simultaneous users you expect.
* You may have 10,000 users taking a test, but only 1 in 4 may be clicking on an answer at any one
time. Therefore there are only 2,500 simultaneous requests.
If your usage is high, a reverse proxy server (eg. squid) in front of the web server will lighten the load on serving
images from the web server if you use a lot of images in your questions.
Using the Communal Question Bank will significantly reduce the amount of space required in
the database and decrease the number of similiar questions.
-------------------------------------------------------------
-------------------------------------------------------------
Apache on Windows:
By default, Apache starts with 1 child process and 1 thread.
** If you are doing testing with default, do not restart the server as it resets the threads to 1, and it will have to load the threads again.
Apache on Windows will create up to 1920 threads as it feels it needs them, but this will take time.
We want a large number of threads (1920) waiting for requests as our initial load is high for a quiz/test.
These threads will not die until you restart the server or kill the child.
In the following text, we kill the child off in case of memory leaks. So a new child is spawned.
We recommend the following text be added to the end of C:\Apache24\conf\http.conf:
ThreadsPerChild 1920
MaxConnectionsPerChild 30000
Save the file and stop then start apache.
------------------------------------------------------------
------------------------------------------------------------
Apache on Ubuntu uses the event_mpm.
By default, the event_mpm begins with 3 child processes and 25 threads. It will max out to 16 child process for a total
of 400 threads.
The threads answer the requests from browsers.
If there are not enough threads to answer the requests, apache will create another child process, then each second will
double the number of children (1-2,2-4,2-8,2-16 up to 32 per second) until it creates enough threads to handle the incoming requests.
On a regular webserver this would usually suffice to handle the increase in requests.
What we find, is the default settings in Apache are usually good for around 30-50 individuals taking a test at the same time.
Any more, and you need to alter the settings to the example we give below in apache2.conf.
With this mpm you can change the children and requests according to your needs and memory available.
If for example you have 50 questions in your test and each question has an image in it and each answer (4 answers) has
an image in it, and all questions are displayed at once, You have 50 x 5 = 250 requests per each users browser when
starting a test. Multiply that by 100 users starting the test at the same time, and the webserver is swamped by 25,000
requests in the first few seconds.
The default settings will not handle this type of load.
With the settings we provide in our install docs (500 threads available) you should be able to easily handle 100+ individual test takers.
** Remember that your network must be able to handle the traffic also.
Any more, and you may need to alter the settings to the example we give below in apache2.conf.
With this mpm you can change the children and requests according to your needs and memory available.
Therefore, with the QST configurations below, we start Apache with 50 children, and each child can handle 100 requests,
for a total ability to handle 5,000 requests concurrently (we are assuming you have 32G of ram and only QST and the database
are running on the server). There is no need to save ram for other uses as only QST is running.
In this scenario, apache will keep creating children as they are needed up to 200 children. 200 children x 100 threads per
child = 20,000 requests. Also, it will always try to have 3000 threads available for requests.
After each child has handled 10,000 requests, it is closed (in case of memory leaks).
We also start with 50 Perl interpreters and each one will have 50 threads, so 2500 threads to answer requests.
----------
Add the following 5 lines to /etc/apache2/apache2.conf below Include ports.conf:
Options FollowSymLinks
AllowOverride None
Require all denied
The above turns off .htaccess lookup which will save the server the expense of checking who can
access files.
Set the following in apache2.conf:
HostnameLookups Off
KeepAlive Off
The following you should change so there are many threads available (5,000) when apache first starts.
StartServers 5
MaxRequestWorkers 10000
ServerLimit 100
ThreadsPerChild 100
ThreadLimit 100
to the following:
StartServers 50
MaxRequestWorkers 20000
ServerLimit 200
ThreadsPerChild 100
ThreadLimit 200
MinSpareThreads 3000
MaxSpareThreads 4000
Save the file, run:
sudo service apache2 stop
sudo service apache2 start
** Do not do a restart, it will not reread apache2.conf!
----------------------
-----------------------
Go to /etc/mysql
Copy mysql.cnf to my.cnf
Put the following into the file:
[mysqld]
max_connections=1000
Save the file.
Restart mysql:
Open /etc/apache2/sites-enabled/000-default.conf
and change to:
PerlInterpStart 50
PerlInterpMax 100
PerlInterpMaxSpare 50
Save the file.
sudo systemctl daemon-reload
sudo systemctl restart mysql
--------------------------
--------------------------
Be sure your internet connection can handle the number of users you are expecting.
Have all your users attempt a quiz at the same time to gauge the load on the server before
doing any important testing to ensure it will handle the load.
Of course you can fine tune the above as required.
Contact us at qstsupport@shaw.ca if you have questions regarding scalability.