Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

System Description

Authors
Affiliations
Cornell University
National Center for Supercomputing Applications
University of Illinois at Urbana-Champaign

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.

SizeProcessorAvailable to your analysisDisk
30 GiB8 cores (AMD EPYC-Milan)≈28 GiB60 GB
60 GiB16 cores≈58 GiB60 GB
125 GiB32 cores≈123 GiB60 GB (by request)
250 GiB64 cores≈248 GiB60 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

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:

Inputs:

TRP: Artifact Filtering

Purpose: Process and filter research artifacts according to .sivacorignore specifications before publication.

Process:

Inputs: