benchmark results #5
Description
Updated Results March 2021
Old Results Sep 2019
Machine information:
Intel(R) Xeon(R) CPU E5-2697A v4 @ 2.60GHz (two sockets)
256GB (32GBx8) DDR4 Synchronous Registered (Buffered) 2400 MT/s
workloads
addr-*:
all addresses in the U.S, Mexico, and Brazil
source: https://openaddresses.io/
az*:
amazon comments. a:productID(asin); r:reviewerID; t:unixReviewTime.
source: https://nijianmo.github.io/amazon/index.html
2014 version. 142 millions keys
dblp-*:
dblp database dump in early 2018.
4-million keys each. Small datasets compared to the others.
m9urls:
memetracker urls dump.
http://www.memetracker.org/data.html
Lots of urls; some URLs are thousands of bytes long.
zero-16m:
16 million randomly generated keys with a lot of zeroes.
HOT vs Wormhole https://github.com/speedskater/hot
The HOT implementation we obtained has several limitations:
(1) char * string key. It does not handle keys that contain zeroes.
(2) key length has a small maximum length (255). We increased it to 2047 to accommodate a few very-long keys in the url keysets.
(3) The multi-threaded version does not support deletion. We only compare the single-threaded HOT with a single-threaded wormhole.
Throughput (mops, Million Operations Per Second) On a Skylake server (updated 2021)
keyset HOT Wormhole
az-art: 1.270 1.802
addr-mx: 1.813 1.819
dblp-title: 1.746 2.501
dblp-url: 2.155 2.506
dblp-ee: 2.329 2.540
m9urls: 0.917 1.155