Capacity Planning - System Architecture for Enterprise 11.5

[Brian_McLeer]

In thinking about our next upgrade from 11.2 to 11.5 (probably to be released around 2025 Esri UC), I am looking for feedback from others. Below is a high-level overview of a proposed production environment. For some background, we are a mid-sized city of about 65-70k. We have about 6-8 admins/SDE editors. We use Field Maps with feature service replicas. We integrate with a CMMS system, a permitting system, and a few others. 

Below is a mix of the current and proposed production environment for our next upgrade, which will be to 11.5 as this is the next long-term support release planned by Esri. Looking for feedback if anyone has any suggestions. It can be that a resource below might be not enough or too much. 

Server Function	Windows Server Edition	SQL Server Edition	RAM	CPU	C Drive (OS and software)	D Drive (Data Storage)	Notes
Database Server	2022	2022	16 GB	4 Cores	100 GB	1 TB
Portal for ArcGIS and ArcGIS Data Store	2022	N/A	16 GB	4 Cores	100 GB	200 GB	Portal and Data Store on same machine
ArcGIS Server (Hosting)	2022	N/A	16 GB	4 Cores	100 GB	100 GB
ArcGIS Server (Federated/Internal)	2022	N/A	24 GB	4 Cores	100 GB	200 GB	Also used to run nightly Python scripts, thinking about migrating to notebook server.
ArcGIS Server (Federated/External)	2022	N/A	48 GB	6 Cores	149 GB	1.34 TB
ArcGIS Monitor	2022	N/A	16 GB	4 Cores	125 GB	N/A

[AngusHooper1]

Leave C to the OS, install software to D and partition out E for data.

Dedicate datastore to its own VM or co-locate with hosting if you have to.

[Brian_McLeer]

Thank you @AngusHooper1, I hadn't thought about a 3-way partition.  

[ChrisBeaudette]

FWIW, we had our datastore co-located on our hosting server and isolated it to its own server at Esri's recommendation (it's part of their recommended architecture). After doing so we started getting issues where any changes to hosted schemas would "lock" that service endpoint. And if users repeatedly attempted to make the schema change it would eventually require a data store restart.  We worked with support for weeks but were not able to find a resolution.  We're now in the process of co-locating our datastore back to our hosting server. We're running on Windows DataCenter 2016 across all servers. Not sure if this is unique to our installation since neither we nor Esri have heard of it occurring to anyone else. We're at ArcGIS Enterprise v11.1 but probably upgrading to v11.5 within a few months.

[DEWright_CA]

I like very much your structure; this matches mine quite well. I would suggest you think about the number of services you plan to be using across your various AGS instances keeping in mind each SOC takes up a standard chunk of RAM for Isolated Pool services, and using Shared Pool has a performance hit.

I would take into consideration if you have any large processing that could put a bigger load on any one set of services; such as large geocoding processing or analysis services that you would potentially want to have that load run on a seperate AGS to not impact the load on your day to day transactional users of your internal or external mapping services.

[Brian_McLeer]

Thank you @DEWright_CA, we have a few hefty map services that contain 100-200 layers for our main internal and external city mapping applications. They can get resource intensive, so bumping up the RAM on both on internal and external federated AGS machines may be something to consider.  

[ChrisBeaudette]

Agreed, in terms of resources required every installation is different in terms of patterns of use.  The best test is to monitor your Production environment over time and adjust as necessary.  If your current infrastructure at v11.2 is handling your load then it should also work for your upgraded environment.

[BillFox]

is a high availability enterprise setup an option for you?

[Brian_McLeer]

@BillFox, I don't think that is an option for us, we are probably too small of an organization to have HA available. We don't have too much downtime, and usually, we can get ArcGIS Security Patches, Windows Updates, etc. accomplished after hours. 

[Lerman]

Hi,

Why are there so few CPU cores? Is it because of license limitations? I don't think they can support concurrent access.