Fix #24 per thread throughput

[ashish (ashyx12)]

Summary

Fixes per-thread throughput is always 0.00 with ntttcp version 5.40.

In v5.40, worker start/end timing was tied to tcp_row, which is only
initialized when TCP EStats are enabled. As a result, normal runs without
-es left worker_time as zero, causing per-thread Time(s) and
Throughput(KB/s) to print as 0.000 even though total throughput was
reported correctly.

This change records worker timing independently from EStats collection while
keeping EStats calls guarded by tcp_row.

Fixes #24 .

Validation

• Built locally in Visual Studio using the ntttcp.sln file.
• Manually ran the reported sender/receiver command pattern.
• Reproduced the issue without -es: per-thread Time(s) and
  Throughput(KB/s) were 0.000 while total throughput was nonzero.
• Confirmed the suspected cause by running with -es: per-thread timing was
  recorded because tcp_row was initialized.
• Confirmed the fix by running without -es: per-thread Time(s) and
  Throughput(KB/s) are now nonzero.
• Confirmed the -es path still reports nonzero per-thread Time(s) and
  Throughput(KB/s) after the fix.

Screenshots

Before: without -es

Per-thread Time(s) and Throughput(KB/s) were reported as 0.000, while
total throughput was still nonzero.

Confirmation: with -es

Running the same command pattern with -es recorded nonzero per-thread time,
confirming that timing was incorrectly coupled to tcp_row.

After fix: without -es

After the fix, the same command pattern works without -es: per-thread
Time(s) and Throughput(KB/s) are nonzero.

After fix: with -es

The -es path still records nonzero per-thread Time(s) and
Throughput(KB/s) after the timing change.

Follow-up for a0a7d6e

This addresses the Copilot code review:

• Added a guard requiring both begin and end EStats snapshots to be valid
  before marking EStats as available.
• Validated the EStats initialization failure path with -es using XML output:
  before the fix, zero-initialized EStats were emitted; after the fix,
  per-thread Time(s) and Throughput(KB/s) remain nonzero and EStats are
  not emitted.
• Confirmed normal -es runs after the fix still emit valid EStats data.
• EStats are marked available only when both the begin and end EStats snapshots were successfully collected; worker timing is independent of EStats collection.

[ashish (ashyx12)]

@microsoft-github-policy-service agree

[Copilot]

Pull request overview

Decouples worker timing from optional TCP EStats collection to restore per-thread throughput reporting.

Changes:

• Records synchronous and asynchronous worker timing without -es.
• Keeps EStats snapshots conditional on tcp_row.
• Marks GetEstats input as optional.

Suppressed comments (1)

src/ntttcp.c:3739

• The asynchronous path has the same EStats failure regression: a null tcp_row no longer prevents time0_was_set, so the exit block marks zero-initialized EStats available whenever -es was requested. Preserve the independent worker timer, but gate estats_available on successful begin/end EStats collection rather than the option flag.

            time0_was_set = TRUE;

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[ashish (ashyx12)]

Guillaume Hetier (@guhetier) i’ve addressed the copilot review in the latest commit, including the EStats failure-path issue, and validated the changes requesting a review when you get a chance.