SIVACOR operates through a series of Trusted Research Performances (TRPs) that ensure computational reproducibility. TRPs are standardized, verifiable processes that form the foundation of transparent computational research. For more information about TRPs, see the TRACE Specification.
Hardware capabilities¶
Every submission runs on its own virtual machine, created when your job starts and destroyed when it finishes. Your code is the only thing running on it, and nothing is carried over from one submission to the next.
You choose the size when you submit — see choosing the machine size. New submissions default to the smallest.
| Size | Processor | Available to your analysis | Disk |
|---|---|---|---|
| 30 GiB | 8 cores (AMD EPYC-Milan) | ≈28 GiB | 60 GB |
| 60 GiB | 16 cores | ≈58 GiB | 60 GB |
| 125 GiB | 32 cores | ≈123 GiB | 60 GB (by request) |
| 250 GiB | 64 cores | ≈248 GiB | 60 GB (by request) |
Your analysis gets all of the cores, and all of the memory except a small reserve — about 2 GiB is held back so that the machine itself, and the agent supervising your run, cannot be starved by the analysis. That reserve is why each “available” figure is lower than the size’s name, and it is an approximation: the exact limit is reported in your run’s performance data.
An analysis that tries to exceed the limit is stopped, and the job log says so and names the figure. It is not silently slowed down or swapped to disk.
Disk is the same at every size. A larger machine buys memory and cores, never space.
The 60 GB of disk is shared between your replication package and the software image it runs in — see Size considerations when preparing your package.
A submission may additionally be given a temporary scratch disk, of a size it asks for, on top of the machine’s own 60 GB. It is granted per account on request, created for the one submission and destroyed with the machine — see extra scratch disk. Nothing about the record of your run changes: the files are in the usual working directory and are hashed into the TRO like any other.
Limits¶
Run time. A run is stopped after 7 days.
Sign of life. A run that reports no activity for 30 minutes is marked as failed, on the assumption that the machine running it has been lost.
Retention. Submissions are deleted 14 days after they are submitted.
Uploads. A single archive may be at most 5 GB, and each user may store up to 10 GB.
One at a time. You can have only one submission in progress at a time.
At busy times, your submission may have to wait for a machine to become available before it starts. See Monitoring Jobs.
TRS: Trusted Research System¶
SIVACOR has the following TRS capabilities:
TRP: Container Execution¶
Purpose: Execute computational research within an container environment.
Process:
Uses a container image from a pre-specified list as the execution environment
Provisions an execution environment
Runs the container, using the user-provided ZIP file and entry point.
Captures execution logs and outputs
May be called multiple times in a chained execution, with each run updating the project space.
Inputs:
Container image reference from the pre-specified list
Execution parameters: network isolation, name of entry point
Resource allocation: the machine size the submission asked for, and any extra scratch disk it was granted — one of each per submission
TRP: Artifact Filtering¶
Purpose: Process and filter research artifacts according to .sivacorignore specifications before publication.
Process:
This is always the last step.
Parses the
.sivacorignorefile from the user-provided ZIP fileApplies pattern matching to identify files and directories to exclude
Filters the output artifacts based on the ignore rules
Preserves the directory structure of remaining files
Inputs:
Project space from all previous TRPs
.sivacorignoreconfiguration file