Threading and Parallelism in EPM

---

Understanding CPU, Core, Thread, and vCPU

Efficient use of computing resources requires clarity on how modern processors are structured and virtualized:

• CPU (Central Processing Unit): The physical processor. A system may have one or more CPUs.

• Core: A processing unit within a CPU that executes instructions. Modern CPUs typically have multiple cores.

• Thread: A stream of execution handled by a core. With hyperthreading, one core can manage multiple threads (usually 2).

• vCPU (Virtual CPU): In virtualized/cloud environments, a vCPU typically maps to a single thread. It represents the smallest unit of compute allocated to a process.

Term	Description	Relation to Others
CPU	Physical chip	Contains multiple cores
Core	Independent compute unit	Can run 1–2 threads
Thread	Stream of execution	Mapped to a vCPU
vCPU	Virtualized thread in the cloud	≈ 1 thread (part of a core)

Note: A vCPU does not represent a full physical core. When a cloud provider offers “4 vCPUs”, this usually means access to 4 hardware threads — not 4 full cores — and therefore you can run 4 parallel tasks, assuming no other bottlenecks.

---

Framework to Define the Right Simulation Setup

Performance is often limited more by available memory than by CPU speed. Follow this step-by-step method to size your simulations appropriately:

1. Determine Available Memory

Start with the total available RAM on your server or VM.

Example:
For the World Bank Planning Team server:
Available RAM: 128 GB

2. Check CPU and Core Information

Identify how many physical CPUs and cores you have.

Example:
1 physical CPU with 4 cores, each supporting 2 threads
→ Total: 8 threads, 8 vCPUs

3. Estimate Memory Usage per Simulation

Run one standard model and monitor its peak memory usage using GAMS Studio:

• Open the .lst file or console output.

• Look for the ProcTreeMemMonitor entry.

• Use the VSS (Virtual Set Size) value — this is the total memory footprint.

Example:
A typical simulation uses 32 GB RAM (VSS value).

4. Calculate Max Parallel Jobs Based on Memory

Divide available RAM by the estimated memory per job:

Max Parallel Simulations = Total RAM / Memory per Simulation

Example:
128 GB / 32 GB = 4 parallel simulations

5. Set Number of Threads per Simulation

Each simulation can then use a subset of threads, depending on your compute layout:

• If your machine has 8 threads total and you run 4 simulations:

• Each simulation can use up to 2 threads

You control this in GAMS using the CPLEX thread option:

option threads = 2;

---

Example Setup Summary

Parameter	Value
Total RAM	128 GB
Per-job RAM usage	32 GB
Max parallel jobs	4
Total threads available	8
Threads per simulation	2
Cores used	4