VM size burst for workshop purposes

[arashnh11 (Arash Nemati Hayati)]

Hi OOD team,

We are planning to host a workshop through OnDemand where we expect ~80-100 users for the session. Per other threads, I am thinking of 8 cores and 22GB memory. (
The VM is using 7GB memory in idle mode. So with your estimation of 150MB per PUN, I am thinking of 15GB additional memory.
the size of data files that might be uploaded would be couple mb so I am hoping the default /tmp root with 20GB storage will server the purpose.
Would you please comment on the above with any additional recommendations needed?

Thanks,
Arash

[jeff.ohrstrom]

Hey! I think the Memory and CPU stats are fine. The /tmp partition, may or may not hold, depending on how your clients upload files. But, the fact that it’s on your mind is good anyhow.

The last thing I’d say is change your apache mpm configs (if you can, if you don’t need PHP libraries, you should). Here are a couple of topics on it. The first details the mpm settings we use, the second says a little bit about where to drop the files (with some formatting). I say this, because everyone is likely going to connect around the same time and that can be a struggle with mpm_prefork.

OOD Host configuration recommendations

Posting general contents of an offline email thread here in case it would be of assistance to anyone else in the future: From: Ashok Ragavendran , Brown University The question relates to server loads and how you manage balancing and provisioning of the actual front end web servers. I am wondering if you can provide any information on the current configuration that is being implemented at the OSC and any metrics on the loads that your experience. Any recommendations you can provide within thi…

MaxRequestWorkers setting

I think here’s a good serverfault topic on the subject. Yes it’s safe to modify these files, but they could be overwritten by a new install, so you’d be wise to write some automation around configuring your apache. Though the file /opt/rh/httpd24/root/etc/httpd/conf.modules.d/00-mpm.conf determines which type of mpm module you use (you’ve linked a CGI config file, that’s not the file/config you’re looking fo). Looks like default is prefork. If you want to switch types, then modify this file, …

[arashnh11 (Arash Nemati Hayati)]

Thanks a lot for the recommendations. Our VM is dedicated to OnDemand so PHP is not needed. Per those threads,
I have modified the following config file for apache to use the mpm_event_event module.

[ondemand server~]# cat /opt/rh/httpd24/root/etc/httpd/conf.modules.d/00-mpm.conf | grep LoadModule
# one of the following LoadModule lines:
# LoadModule mpm_prefork_module modules/mod_mpm_prefork.so
# LoadModule mpm_worker_module modules/mod_mpm_worker.so
LoadModule mpm_event_module modules/mod_mpm_event.so

I also added worker.conf file to the httpd/conf.d files.
[ondemand server~]# cat /opt/rh/httpd24/root/etc/httpd/conf.d/worker.conf

<IfModule mpm_prefork_module>
MaxRequestWorkers 512
</IfModule>
<IfModule mpm_event_module>
</IfModule>
<IfModule mpm_worker_module>
</IfModule>

and the did the updates:
/opt/ood/ood-portal-generator/sbin/update_ood_portal
systemctl try-restart httpd24-httpd.service httpd24-htcacheclean.service

Is there a way to check that the new mpm module is actually being used by OOD?
Do you have recommendations for what to put under mpm_event_module based on the number of PUNs? or hardware configs?

Also, I recall reaching to almost 8GB RAM with one user and many instances ~ 100. Would that behavior change with the mpm event?
Seems like mpm_prefork is RAM memory thirsty?

[tdockendorf (Trey Dockendorf)]

I don’t know of a way to check MPM is working other than checking mod_status web URL if you have that enabled. I recall the output and metrics displayed will change with MPM event instead of prefork. I don’t recall what metrics change but I do recall the metrics do change as I ran into that when parsing the output. There must be at a MPM enabled and I believe you can only have one so if Apache starts and works then it’s working.

Threads are usually less resource expensive than processes so I would expect a thread based MPM to perform better than a process based one. The MPM likely won’t impact the resources used by the actual PUN from OnDemand but rather the resource consumption of the Apache processes.