Tuning observium

We use observium enterprise 25.9.14293 . Observium is installed on a Virtual machine.

CPU is :

# lscpuArchitecture:                         x86_64CPU op-mode(s):                       32-bit, 64-bitByte Order:                           Little EndianAddress sizes:                        46 bits physical, 48 bits virtualCPU(s):                               6On-line CPU(s) list:                  0-5Thread(s) per core:                   1Core(s) per socket:                   6Socket(s):                            1NUMA node(s):                         1Vendor ID:                            GenuineIntelCPU family:                           6Model:                                85Model name:                           Intel(R) Xeon(R) Gold 6128 CPU @ 3.40GHzStepping:                             4CPU MHz:                              3392.030BogoMIPS:                             6784.06Hypervisor vendor:                    MicrosoftVirtualization type:                  fullL1d cache:                            192 KiBL1i cache:                            192 KiBL2 cache:                             6 MiBL3 cache:                             115.5 MiB

Memory :

#free -tmh               total        used        free      shared  buff/cache   availableMem:            15Gi       3.5Gi       891Mi       653Mi        11Gi        11GiSwap:             0B          0B          0BTotal:          15Gi       3.5Gi       891Mi

We would like to optimize it to speed up page loading, especially on weekends when it is slow.

We have ~23,000 ports and 358 devices.

I have several questions:

1. What is a large installation? I often read this term, but I don't know how to measure or define it.
2. How can we speed up the loading time of pages and graphs?
   1.  We have enabled "opcode caching."
   2. We already enabled "fast userspace caching."
   3. Disable MySQL binary logging done

1. I attached a screenshot of the "pollerlog" page. I read in the tuning section of the website that :

" Ideally, the entire Poller-wrapper process should take as close to 300 seconds as possible to ensure the lowest average load".

If I understand correctly, we have doubled that. How can we optimize this? Is it thanks to "multiple poller instances"?

4. Also, I noticed that the polling time is not good when the equipment is far from where Observium is located (FR). In this case, would it be a good idea to increase the "SNMP Maximum Repetitions"?

5. Finally, I'm not sure if this is a good idea, but we could create a poller dedicated to each country. For example, Poller 1 could monitor all hardware in India, Poller 2 could monitor all hardware in the US, Poller 3 could monitor all hardware in Canada, and Poller 4 could monitor all hardware in France.

6. In general, is it time to consider horizontal scaling?  Front-end and mysql on the same VM and pollers on multiples VM.

Quick question: Is the trap still unsupported? I read an old post. I'm asking because we would like to retrieve the status of an SRX cluster, but Juniper only provides one way to do that: traps. We could use Nagios or another technology, but we would prefer to avoid using multiple monitoring technologies for our network equipment.

Thanks you.