Increase threads versus adding MID servers

rgu
Tera Contributor

From the perspective of trying to speed up discovery scans, has anyone experimented with increasing the threads above the default of 25?

Currently our MID servers are set to 25 threads. When running discovery scans the servers are basically pegged. It is not clear to me what benefit there would be to increasing the threads.

From your experience, what are the bottlenecks on MID servers when running scans? CPU? Cores? RAM? Disk speed? Our MID servers are 2 CPU with 2 GB of RAM.

Besides adding additional MID servers, have you found other tweaks or strategies to speed up discovery scans?

(I did search the forum before posting this question. There were no recent discussions of this topic and/or details specific to Geneva.)

Thanks for your time.

9 REPLIES 9

Just for thoroughness, go validate in the wrapper-override file in the same directory - at the very bottom - make sure that it does not have a pound sign:


#wrapper.java.maxmemry=1024


aleck_lin
ServiceNow Employee
ServiceNow Employee

Make sure you do not modify the wrapper.conf, always modify the wrapper.override file.



If cluster is not part of your concern, on a purely MID server basis, I definitely have seen folks increase the thread from 25 to 75 or 100. Make sure that you increase the Java memory accordingly. Our currently default has been increased from 512MB to 1GB, so if you're going to change the thread to 100, it only stands to reason that you need to change it to the 4GB. Obviously the overall memory on the server should be increased as well; mathematically speaking, you should probably only need 4GB for MID application and maybe 1-2 GB for the rest, but no one would have 6GB of RAM, that's just weird; so 8GB it is.



Also, the speed will depend on your processing power. I would pay attention to the CPU utilization during Discovery. You definitely don't want to increase the threads/memory if the CPU is already pegged at 100%.



Make sense I hope?


Can you please the steps how to increase the threads.

 

rehani
Tera Contributor

On one of the recent webinar presentations by ServiceNow it was mentioned the MID Servers should really have 8GB of RAM up from their existing recommendations.   It was a comment made by the presenter rather than appearing on a PowerPoint.   I've looked at so many recently I don't remember the exact one but its likely a Helsinki or Geneva based webinar on Service Mapping or Discovery.


johnnyjava
Kilo Guru

According to the following document, MID Servers should have 4GB per "application", which I take to mean agent / instance.



MID Server system requirements



When adjusting your VM ram for a MID Server host, remember that your OS needs some of that too.



Steve did a nice write up over here ServiceNow Discovery 101: Virus Scanners And MID Server Performance which might also help you out.



I would strongly suggest getting clustering working, for many reasons.



Also if you had enabled DEBUG logging on your Discovery during implementation, I would suggest you disable that stuff too. Logging (of any sort) is costly in terms of performance. (for the MID obviously, but also anywhere else.)