cqy123456
cqy123456
The IO performance of the test machine is 23x better than AWS's 1cu. And default global search thread pool size of knowhere is (4 * cpu). If IOPS is too...
When HNSW use random dataset, it will often happen early stop search iteration and return.
todo: Scalar type implementation of mmap is incomplete，it can not just simple use a ptr to represent. 1. vachar: use typedef std::basic_string; 2. Array: class member:data_ use mmap allocate; 3....
/run-cpu-e2e
Growing index and Binlog index share same knowhere build index pool. The pool size is 0.5 * cpu number. When push too much tasks into the pool, load segment cost...
After setting knowhere thread pool size by pod resource, build time between 2.2.0 and master is close in knowhere. 2.2.0:  master: 
/run-cpu-e2e