Load balancing challenges

Hello PPP, 

Some things to check around the traffic patterns:

1. Are both publishers egressing using the same firewall Gateway public IP? If yes, they should be ideally connecting to the same Netskope data center based on the egress IP. 

2. Are both publishers configured in the same subnet? If yes then ideally they should have the same path to the private application. If traffic is heavily inclined to one publisher, this makes me think are they in different subnets so they have a notable latency difference to the same private app. 

3. Clients can be inside the office or Roaming users can connect to any DC which you cannot influence directly. 

If still you see the problem, please take MTR traces from both publishers and evaluate the egress path to Netskope DCs and also to the private app. This can help clarify things better.  For any technical evaluation, you can also log a case with support, but please have the traces attached. 

Please let me know in case of any questions. 

BR,
Hitesh 

Hello,

thank you for your answer.

Feel free to correct me if I’m wrong at any point. 

This is the description for NetSkope client but from my understanding there is a similar mechanism for publishers:

• The Netskope Client connects to Netskope’s API (gateway.gslb.goskope.com) using HTTPS (tcp/443) to request a list of nearby Netskope data centers.
• Netskope’s API uses the public IP of the API request to look up the geolocation of the endpoint.
• Netskope’s API then responds with a list of geographically nearby Netskope data centers.
• The Netskope Client tests latency to each of the nearby Netskope data centers.
• The Netskope Client utilizes these latency measurements to pick the best data center (DC). This is generally the one with the lowest latency (RTT); however, in some cases a nearby DC with similar latency characteristics may be selected.

Is there a mechanism that selects the same data center for the same external IP address, or is it based on RTT, which can vary depending on the time of day, especially when the difference is in the range of 1–2 ms? I think there's an equivalent class, so several DCs are taken into account. 

Is there a specific list detailing how a DC is selected (RTT, geo, egress IP, equivalent class)?

From what we can see in the log, the time measurement takes place from the client's gateway to the publisher (via the Stitcher). I don't see the part from the publisher to the actual server here.

US-LAX2 (Gateway Location) → Publisher (Latency range: 64-127 ms)