Have a question regarding the balancing of pathsend operations. We have multiple single process serverclasses which are doing pathsend operations to a serverclass which has four static processes (no dynamic) each running in its own CPU.

Originally setup these destination serverclass processes are configured to handle 1/4 of the load and this was assuming the load would be spread somewhat evenly across the processes. However we are finding loads similar to the current production snapshot: CPU 0 14% - CPU 1 23% - 34% - 29%. During the day these percentages will shift to, for example: CPU 0 29%- CPU 1 34%-23%-12%. Initially I thought the load was due to the CPU that the senders were located on, but once I found that the load was distributed differently depending on when I took a snapshot, the senders never move CPU so can't be the cause.

Is there a way to balance this from pathway perspective or totally out of our control?

On Wed, 23 Feb 2022 11:20:04 -0800 (PST), SRSeedBurners <sterlingranch@gmail.com> wrote:

>Have a question regarding the balancing of pathsend operations. We have multiple single process serverclasses which are doing pathsend operations to a serverclass which has four static processes (no dynamic) each running in its own CPU.
>
>Originally setup these destination serverclass processes are configured to handle 1/4 of the load and this was assuming the load would be spread somewhat evenly across the processes. However we are finding loads similar to the current production snapshot: CPU 0 14% - CPU 1 23% - 34% - 29%. During the day these percentages will shift to, for example: CPU 0 29%- CPU 1 34%-23%-12%. Initially I thought the load was due to the CPU that the senders were located on, but once I found that the load was distributed differently depending on when I took a snapshot, the senders never move CPU so can't be the cause.
>
>Is there a way to balance this from pathway perspective or totally out of our control?

Your source Pathway servers get granted links to the target serverclass on their first call to SERVERCLASS_SEND_.
This is not done on a load-related manner, it is done based on current available static links and other properties (like linkdepth etc.)

Unless there is an error on that granted link, the source server will ALWAYS send to the same target server process, until a request takes longer than 'CREATEDELAY' to complete.
You have no direct control over this, the only control is in managing CREATEDELAY, DELETEDELAY, MAXLINKS and LINKDEPTH (and NUMSTATIC to an extent).

Bill

On Wednesday, February 23, 2022 at 2:20:06 p.m. UTC-5, SRSeedBurners wrote:
> Have a question regarding the balancing of pathsend operations. We have multiple single process serverclasses which are doing pathsend operations to a serverclass which has four static processes (no dynamic) each running in its own CPU.
>
> Originally setup these destination serverclass processes are configured to handle 1/4 of the load and this was assuming the load would be spread somewhat evenly across the processes. However we are finding loads similar to the current production snapshot: CPU 0 14% - CPU 1 23% - 34% - 29%. During the day these percentages will shift to, for example: CPU 0 29%- CPU 1 34%-23%-12%. Initially I thought the load was due to the CPU that the senders were located on, but once I found that the load was distributed differently depending on when I took a snapshot, the senders never move CPU so can't be the cause.
>
> Is there a way to balance this from pathway perspective or totally out of our control?

Consider looking into Pathway ACS. You can balance across PATHMON processes based on percentages. This sits on top of what Bill described regarding links. Look this up in the Pathway ACS manual.