cancel
Showing results for 
Search instead for 
Did you mean: 
Create Post
Level 10

Additional hardware for stacked engines?

I thought I saw somewhere that when you add an additional engine to your main poller you should also increase the hardware platform.  Now I can't find that information, although it sounds logical.  How much burden does the additional engine put on the server?

We use SAM heavily and are reaching the point where polling goes unknown and we start having lost jobs.  I have several hundred servers which are only being polled with NPM because the system cannot handle the application monitoring that needs to be applied to them and several hundred more servers I will be adding in the future.  We have purchased the additional polling engines and I have a plan to build a new server with two engines.  I just want to make sure I spec out the hardware appropriately because adding resources after the fact is a tedious process for us.

Currently I have SAM, NPM, NCM, and SRM.  Soon I will have IPAM installed as well.

My server has the following: it's a VM

Memory 32GB

CPU - 6

HD - 100GB

DB Server

Memory 80GB

CPU- 4

HD - 140GB + 50GB additional share.

Polling Engine statistics:

Elements = 6484

Network Node Elements = 628

Volume Elements = 3107

Interface Elements = 2749

SAM Polling rate = 52% When this reaches 80% or above we experience resource failures and timeouts.

Polling rate = 60%

VM Hyper-V polling rate = 50%

Total Job Weight = 4365

Pollers = 24907

thank you,

Jan

0 Kudos
8 Replies
MVP
MVP

This article is over 2 years old now. Did try to search for some guidelines around System requirements for Stackable Pollers, could not find any documents around that.

We have 4 Additional Pollers each with 8 vCPU's and 16 GB RAM. We are in the need to do a HA solution for the Additional Pollers too ..

A simple way would be to buy 4 licenses and create 4 HA Pollers.

Alternate way would be create a "LARGE" APE and stack all 4 Pollers on the LARGE System and create just one HA Server.

The question is "How LARGE is large"? 32 vCPU's and 64 GB?

abdhijasharmagangadhar.ksagar.b99kushaldevendrab

0 Kudos

From my experience it is a bit vague because the workloads can vary pretty dramatically when you are running quad stacked pollers.  I've definitely worked with clients who heavily use stacking and they were using 24 cpu/32 gb ram and had lots of capacity to spare, but your environment and needs may vary.

- Marc Netterfield, Github
0 Kudos

There has been a big discussion about that here: https://thwack.solarwinds.com/docs/DOC-190008

My personal opinion about stacked polling engines is also in that thread so I am not going to rant here as well.

just like mbird​ mentioned,  I would not stack polling engines either but have separate vms. you could also utilize the hardware more efficient by looking on that post here Tip: How to Improve Orion Performance by Splitting the Module Engine to Separate Processes

Is your SQL box handling the Load with only 4 CPUs, that's really impressive, most of my clients have at least 12 Cores in their SQL Boxes.

i also suggest taking off the webserver load by purchasing an additional webserver. It has the best value per Euro/Dollar/%putyourcurrencyhere% of all licenses (just my opinion)

0 Kudos

If it were up to me I'd like to be able to pull the web server off the primary poller without having to pay for another license, but it's just not like that unfortunately.

Level 12

I have 3 APEs, none of them are stacked so they each are separate VMs. Each VM has 4x CPU, and 24 GB of memory, and I do not have any problems polling.

Current environment stats are:

NETWORK ELEMENTS 13452

NODES 1076

INTERFACES 9679

VOLUMES 2697

And about 37% on the VMWare polling.

0 Kudos

That's a little different scenario as mine will be on the same server, not split between two, and technically I have three independent SolarWinds servers, each will have two engines.  SAM is the big dog on the server, the burden it puts on the engines is intense.  

0 Kudos

Honestly I'd imagine you'd be OK if you stacked two on one VM, but stacking one on top of the primary might get hairy if you have a lot of elements. Even with the load distributed I still see the primary getting hammered a bit as it also has to run the web console, too.

0 Kudos

The idea is to evenly distribute the elements between the two engines to reduce the burden.   We're not close to the count limitation on the engine, which is why I am concerned with possible resource contention.    We had issues in the past with polling and we added more cpu/memory and it resolved the issue.   I thought I read to double the hardware resources when adding the additional engine.

0 Kudos