I think (though not from first-hand experience) there may be some ugliness in keeping large binary data in git ? If the data is reasonably small, then I would

I think (though not from first-hand experience) there may be some ugliness in keeping large binary data in git ? If the data is reasonably small, then I would pick S3 (or some other redundant cloud object-storage) as an easy and disaster-proof solution. If it gets big, then access bandwidth and latency and cost may all be a problem. Maybe a master copy in S3, and a copy on hard disk of some machine near the benchmark system ?