ArcGIS Server 10.7.1: What is the minimum network capacity (bandwidth) in Mbps that ensures stability for machines joined in same site?

1605
4
09-20-2020 01:45 AM
JamalNUMAN
Legendary Contributor

ArcGIS Server 10.7.1: What is the minimum network capacity (bandwidth) in Mbps that ensures stability for VM servers joined to same site?

 

I couldn’t figure out What is the minimum network capacity (bandwidth) in Mbps that ensures stability for VM servers joined to same site knowing that the number of requests reach 160 request/hour

----------------------------------------
Jamal Numan
Geomolg Geoportal for Spatial Information
Ramallah, West Bank, Palestine
4 Replies
DavidHoy
Esri Contributor

Hi Jamal,

The stats graph you have provided shows a peak of 80,000 in the busiest half hour - which is not enormous - but still sizable. that's probably something like 64 per second.  If you have 3 x 4 vCore machines in your site, that means you need an average request serviced in 0.2 second to maintain that peak continuously (that's certainly achievable with modern CPU) but may be affected if your services are not well optimized (scale suppression, geometry simplified etc.)

The network throughput required is mostly dependent on the size of the response packets - worst case, you are returning rendered map pages to a desktop client - that could be maybe 1 MB per request. - but most likely the majority of the requests are Map Service queries or Feature Service retrievals - and these are generally small < 1K each.

so - worst case, 64 MBps = 520 Mbps

best case, 64 kBps = 0.5 Mbps

somewhere in between is the answer, but practically, a modern data centre will be providing minimum 100 Mbps NIC and more likely 1 Gbps.

If you are wondering about inter-machine chatter, there is none, the load balancing is done up the chain - in the Web Adaptor(s) and/or Load Balancer - and the server bandwidth requirement is shared across all the machines in the site.

JamalNUMAN
Legendary Contributor

Thank you David for the valuable input.

 

I couldn’t figure out how the calculations went on this

 

the maximum rate is 80,000 request/30 minutes = 45 request/second

 

assuming that 1 request = 1 MB, then

 

45 request/second = 45 MB/second = 450 Mbps = 0.45 Gbps <1Gbps

 

But how the request is assumed to be 1MB?

----------------------------------------
Jamal Numan
Geomolg Geoportal for Spatial Information
Ramallah, West Bank, Palestine
0 Kudos
DavidHoy
Esri Contributor

Jamal NUMAN

1 MB is a pretty raw and rough guess (based on benchmarks run some years ago) at the  amount of data that is transferred from data storage (database or file system) to a desktop client - but would also apply for INTERNAL traffic (DB to Server) to a Server instance prior to the service request response being pushed out to the calling client's app/browser. The amount of data sent across the web is generally significantly less than the raw data => a Map export may be 100-1000k depending on complexity of the image but a feature service .json file or the response to a "query" on a Map Service probably much less.

DavidHoy
Esri Contributor

Jamal NUMAN

I just wanted to check something else I picked up from your screenshot of Server Manager.

You have three machines as members of your one GIS Server site

Orthophotos, Parcels & UPMS. - all behind one Web Adaptor called "ad" I presume.

I presume you are aware that in the one site, all the machines will be running all of the services published by the site.

The web adaptor is responsible for load balancing the incoming requests to the individual machines and it does this on a round-robin basis.

You cannot allocate (say) "orthophotos" services to just the one "orthophotos" machine.

So - your 45 requests per second are being spread across three machines - so down to 15 requests per second per machine. The NIC of the individual machine is carrying less traffic.

The NIC for the web adaptor and outward bound traffic would be carrying the full load.

But if your underlying question is "is a 1 Gbps" NIC adequate?" - absolutely yes.

Any network might become unstable at over 50% utilization for sustained times, but I dont see you coming anywhere near that.