We currently download the entire VEP cache of interest into every VM/instance, which takes a significant amount of time. When running smaller test sets, the downloading and unzipping likely accounts for most of the "annotation" time. gcsfuse might be able to alieviate this problem.
On the other hand, there may be some relative latency per variant, which could accumulate significantly on larger data sets. Performance testing would have to be done.
We currently download the entire VEP cache of interest into every VM/instance, which takes a significant amount of time. When running smaller test sets, the downloading and unzipping likely accounts for most of the "annotation" time. gcsfuse might be able to alieviate this problem.
On the other hand, there may be some relative latency per variant, which could accumulate significantly on larger data sets. Performance testing would have to be done.