CCSONP - NO DISPENSE IF MOD PAIR (or Mod) TOO BUSY

It's been years since “ALL” TPF MOD PAIRs (P&D) have produced the same performance and this disparity is getting worse with FICON; due to multiple device activity per channel, hundreds of devices per frame, tens of channels per CPU and tens of CPUs, each CPU having awesome speed - Every mod pair works differently.
Mod pair and Device overload is / has become common: To some manageable; causing others to add their own reducing agents, while others just wait.

When “OVERLOAD HITS”- it's often too late and just like sharks at feeding time – so too is the reaction; some vague decreasing / delaying agent is added; Thus making TPF's speed slower.

Isn't “SLOWING TPF” the worst thing one can do?

What's helpful is a SIP selectable option in which users avoid OVERLOAD to purchase reaction time.

This SIP option is:

STOP dispensing NEW POOL to a STRESSED OUT mod of a pair.

Look for and use non-stressed Pairs.

Maintain TPFs speed as long as possible, when possible.

The number of mods used by TPF will grow and growth is good for TPF – growth should be more speed.


Mod Pairs that “do more” should get more.

The pairs who do less:

These need to avoid catastrophic processing, away to do that is to remove Pool Dispensing.

No dispense drops IO activity by more than 40%.

The slow mods have an easier time of staying even until a fix is added.

Thus allowing TPF to maintain speed.

later - Lovely monitoring and managing tools can be added.


CCSONP already has no pool dispense for MOD down - this processing can be expanded to include STRESS conditions, like:

High Mod Queue,
Mod Copy,
Mod Capture,
Mod Mount / Dismount,
CC3 processing,
Path down & Errors,
etc.