I saw #NCBI 's post about the #BLAST scaling problem. What if we replaced the #LMDB backend with an in memory database like #Redis? This could reduce latency by avoiding Disk I/O and improve scalability through sharding/threading. Not sure how big the current dataset is/will be but #aws offers EC2 instances with up to 24 TiB of RAM engineered specifically for low latency.

Is LMDB the bottleneck, or is the corpus of nt's just too large to search quickly? Would love to hear thoughts on this