search

carbynestack / klyshko

Carbyne Stack correlated randomness generation subsystem
Apache License 2.0
4 stars 6 forks source link

Tupletype-specifc thresholds and job sizes #69

Closed strieflin closed 1 year ago

strieflin commented 1 year ago

For cryptographic offline phases, the throughput for different tuple types can differ wildly. E.g. the throughput for the Cow-Gear CRG and GF2N is two orders of magnitude slower than for GFp. With a single threshold and job size (= number of tuples per job), one either has to wait for very long times until all tuples are available or has to use very small jobs which results in high overhead.

The proposal is to make both the threshold and the job size configurable. The most suitable places for this probably the scheduler CRD.

strieflin commented 1 year ago

Started working on this in #70.

strieflin commented 1 year ago

This has been implemented in #70.