[Opensearch] Vector Search ๋ฒค์น˜๋งˆํฌ (vs Elasticsearch)

[์›๋ณธ ๋งํฌ]

๊ด€๋ จ ๋ฒค์น˜๋งˆํฌ
https://blog.naver.com/sssang97/223840387576
https://blog.naver.com/sssang97/223848053614

1 CPU, 8GB ๋ฉ”๋ชจ๋ฆฌ ์ •๋„๋งŒ ์ฃผ๊ณ  256์งœ๋ฆฌ ๋ฒกํ„ฐ 1000๋งŒ๊ฐœ๋ฅผ ๋ฐ€์–ด๋„ฃ์–ด์„œ ๊ฐ„๋‹จํ•œ ๋ฒค์น˜๋งˆํฌ๋ฅผ ๋Œ๋ ค๋ดค๋‹ค.

๋„ฃ๋Š”๋ฐ๋Š” 19์‹œ๊ฐ„ ์ •๋„๊ฐ€ ๊ฑธ๋ ธ๋‹ค.

๋„ฃ๊ณ  ๋‚˜์„œ ๋ณด๋‹ˆ, ์‚ฌ์šฉ ๊ธฐ๋ฐ˜ ์บ์‹œ์— ์˜์กดํ•˜๋Š” Elasticsearch ๋ณ€์ข…๋‹ต๊ฒŒ ์ตœ์ดˆ ๋ ˆ์ดํ„ด์‹œ๋Š” 1์ดˆ๋ฅผ ์ดˆ๊ณผํ•˜๊ธฐ๋„ ํ•˜๋Š” ๋“ฑ, ๊ทธ๋‹ค์ง€ ๋น ๋ฅด์ง€๋Š” ์•Š์œผ๋‚˜, ๋ฐ˜๋ณต๋˜๋Š” ์ฟผ๋ฆฌ์— ๋Œ€ํ•ด์„œ๋Š” ๊ฝค ํ›Œ๋ฅญํ•œ ์„ฑ๋Šฅ์„ ๋ณด์—ฌ์ฃผ๋Š” ํŽธ์ด์—ˆ๋‹ค.

ํ‰๊ท , ์ตœ๋Œ€, ์ตœ์†Œ ์ฒ˜๋ฆฌ์†๋„๋Š” ์ƒ๊ฐ๋ณด๋‹ค ์ค€์ˆ˜ํ–ˆ๋‹ค.

Elasticsearch 8.*๋ฒˆ๋Œ€๊ฐ€ 1์ดˆ๋Œ€ ๋ฐฉ์–ด๋„ ์ž˜ ๋ชปํ–ˆ๋‹ค๋Š”๊ฒƒ์„ ์ƒ๊ฐํ•˜๋ฉด ๋งค์šฐ ๋›ฐ์–ด๋‚œ ์ฐจ์ด๋‹ค.

๊ทผ๋ฐ ์–˜๋งŒ ๋ฒ„์ „ ์ตœ์‹ ์œผ๋กœ ํ•˜๋ฉด ๋ญํ•˜๋‹ˆ๊นŒ Elasticsearch 9.*์œผ๋กœ๋„ ํ•ด๋ดค๋Š”๋ฐ

๊ทธ๋ž˜๋„ ์ข€ ์ฐจ์ด๊ฐ€ ๋‚˜๊ธด ํ–ˆ๋‹ค.

Vector Search๋งŒ์„ ์›ํ•œ๋‹ค๋ฉด ์„ฑ๋Šฅ์€ ํ™•์‹คํžˆ Elasticsearch๋ณด๋‹ค Opensearch๊ฐ€ ํ›จ์”ฌ ๋‚˜์€ ๊ฒƒ ๊ฐ™๋‹ค.
๊ทธ๋ž˜.. ๋Œ€๊ธฐ์—…๊ณผ ์‹ธ์šฐ๋ฉด ์ด๊ธฐ๊ธฐ ์–ด๋ ต์ง€... ์™œ ๊ทธ๋Ÿฐ ์ง“์„ ํ•ด์„œ.



https://github.com/myyrakle/benchmarks/blob/master/vector_search/src/bench_opensearch.rs