Skip to main content
Simulator software 4.7.0 · Released 2026-09-08 · Supersedes release 0.1.1292

​
At a glance

This release extends the platform’s optimized execution path to simulations submitted through the full API, at up to 4,096 nodes. It also makes the RoCE NIC segment size configurable — enabling 9K MTU studies — improves start-up reliability for the largest tracesets, and includes security updates.
No changes in this release alter results for existing configurations run unmodified. All newly exposed attributes default to their previous effective behaviour, and execution-path selection does not change simulated results. Runs on this release are directly comparable with runs on 0.1.1292.

​
Optimized execution path now serves the full API

On earlier releases, the optimized execution path — which substantially reduces end-to-end simulation time — was available only to simulations submitted through the basic API. Simulations submitted through the full API always ran on the standard path, regardless of size. Eligibility is now decided per launch, from properties of the launch itself rather than from how it was submitted:
  • Size — configurations of up to 4,096 nodes qualify.
  • Trace inputs — the launch’s trace-input form must be supported by the optimized path, including zip trace archives.
  • Chakra ID mapping — a rank mapping must be present.
A launch that meets all conditions runs on the optimized path automatically; anything that does not is served on the standard path exactly as before. Nothing that ran previously is refused, and no configuration change is needed to opt in — qualifying full-API launches simply run faster after the upgrade. Two things to be aware of:
  • Path selection does not affect results. The same simulator binary runs in both cases; only orchestration and end-to-end turnaround change.
  • Each launch records why it was routed the way it was. If a run you expected on the optimized path ran on the standard path, the recorded reason identifies the first condition that did not hold — useful when raising the question with your Scala representative.
Related to large launches, container start-up now runs under a timeout sized to the data being loaded. Previously the largest tracesets could fail at start with a timeout error before the simulation began; see Platform reliability.

​
New in this release

​
ScalaRoceNic model 4.5.1

The RoCE NIC’s maximum segment size is now configurable. The MSS used to segment WQEs into data packets was fixed at 4096 bytes, which made 9K MTU studies impossible to represent. It is now exposed as ScalaRoceQpManager.MSS: The default is unchanged, so existing configurations behave exactly as before. Out-of-range values are rejected at configuration time with a specific message, not at run time. In the same model revision, the MRRS attribute has been hidden: the model never reads it, and a visible control that cannot affect a result is a defect. Model documentation is updated accordingly.

​
Zip trace archives

Uploaded .zip tracesets were not tagged with their format on completion, which could misroute them at launch. Zip archives are now recognized as Chakra archives end to end, and a traceset mixing zip and non-zip content is served on the standard path rather than failing.

​
Model catalog changes

  • ScalaRoceNic 4.5.1 is published, with the MSS attribute described above included in the shipped default templates.
  • The shipped default templates no longer list HdrmAllocSfMgn and XonOffset for ScalaSwitch. These attributes were hidden in model 3.6.0 because they could not affect any result; this release removes them from the templates as well.

​
Platform reliability and stability

  • Large tracesets no longer time out at container start. The container start step previously ran under a fixed internal timeout that was too small for the largest tracesets in the catalogue, failing runs with a timeout error before simulation began. The timeout is now sized to the data being loaded.
  • Terminal failures report the underlying cause. A failed simulation previously reported only a coarse failure category; the specific error explaining why was lost with the compute instance. That detail is now captured and carried back, so failures can be diagnosed precisely and without rerunning.

​
Security updates

This release contains security updates. We recommend upgrading. In keeping with our disclosure practice, we do not publish details of security fixes. Customers with a specific compliance question should contact their Scala representative.

​
Documentation

RoCE NIC model documentation has been updated for the newly configurable segment size, including the attribute’s interaction with link MTU and the per-packet header budget.

​
Not included in this release

The following remain open and are not addressed by 0.1.1427:
  • Termination of simulations that are not in a running state. Termination currently applies only to running simulations.
  • Pre-simulation run-time estimation. There is currently no supported way to determine in advance whether a trace will exceed the configured simulation time limit.
  • ECN-marked packet counts received at the NIC. NIC-side receive counters for ECN-marked packets are not yet reported. Switch-side ECN information remains available.

​
Upgrade notes

  • Model catalog updates apply automatically when the platform is upgraded. No action is required.
  • Existing configurations continue to work and behave identically. MSS defaults to its previous fixed value.
  • Qualifying full-API simulations will run on the optimized execution path automatically after the upgrade. Results are unchanged; end-to-end times for those runs will drop.