luyuncheng

> 1. Is the file size a good estimate of the memory that the graph will take up? The plugin uses the file size as an estimator of memory i...

> Could you run benchmarks for 3 different data sets like you ran in #945 @jmazanec15 OK, give me some time, i'll benchmark it. > could you provide more details...

@jmazanec15 i did 1st benchmark with dataset `gist-960 `, and next i will do other dataset: glove-200 , sift-128 Dataset: gist-960 Machine: 8Core16G, client + node in one single machine....

> Could you also add p50 and p90 query metrics as well? @jmazanec15 Updated, AND also i added R:96 in NSG graph, it shows when R increase linearly, query took...

i did 2nd benchmark with dataset `sift-128 1M `, next i want to do some specific memory usage tests Dataset: sift-128 1M Machine: 8Core16G, client + node in one single...

> indexing seems to be more expensive for NSG, correct? Query times do look better though. @jmazanec15 Yes, correctly. it tooks more time to create NSG grpah at index time....

> similarly for M=16, for sift, this value would be 0.65. > For R=64, it would be 0.71 GB. @jmazanec15 Thanks for the formula for memory estimate. i am curious...

> @luyuncheng I see thanks for providing that Ill take a look at #1139 as well. Overall, memory limit seems pretty good. Im wondering if for querying they use 32-bit...

> Also, for HNSW test, did you use nmslib or faiss? @jmazanec15 i think this caused by `malloc_trim(0);` at latest commits, i removed all `malloc_trim(0);` calling. and some times it...

@jmazanec15 i think there is some memory usage optimize as https://github.com/opensearch-project/k-NN/issues/946#issuecomment-1725658197 shows, also #1139 can help me evaluation the memory usage. and i found some memory usage waste. so i...