This message was deleted.
# general
s
This message was deleted.
s
It depends on how you planning to access Venice database. If you are planning to run it natively using DaVinci Client which loads the database locally in the accessing host, you could expect sub-millis performance, For remote queries using venice-thin-client, singlet gets would be < 10ms.
Regarding all-in-RAM datasets, if you enable PlainTable format in RocksDB you could expect orders of magnitude performance gain. In DaVinci client it would be similar to access patterns of RAM vs disk (or SSD).
c
Thanks!
m
The 10ms p99 latency is largely due to the router layer and GC issues. We also have a fast client which talks directly to servers skipping routers giving 2ms P99 latency in our workloads.
The 10ms and 2ms are our internal SLO guarantee numbers the actual numbers are way smaller in most cases.