AMD EPYC vs. Intel Xeon Cascadelake With Facebook's RocksDB Database

Following the benchmarks earlier this month looking at PostgreSQL 12.0 on AMD EPYC Rome versus Intel Xeon Cascade Lake there was interest from Phoronix readers in wondering how well Rome is doing for other modern enterprise database workloads. One of those workloads that was recently added to the Phoronix Test Suite / OpenBenchmarking.org is Facebook's RocksDB, the company's embedded database that is forked from Google LevelDB. With RocksDB being designed to exploit many CPU cores and modern SSD storage, here are some benchmarks looking at how the Xeon Platinum 8280 stacks up against various new AMD EPYC 7002 series processors.

RocksDB is a key-value embedded database solution that Facebook has been working on since 2012 in taking Google's LevelDB to the next level of performance on modern CPU/SSD servers. RocksDB is in turn also used by companies like LinkedIn, Airbnb, Pinterest, Rakuten, Uber, and others.

With RocksDB having its own performance-focused built-in benchmarks, it makes for some interesting performance comparisons on these server CPUs given its growing presence in the enterprise. Those unfamiliar with RocksDB can learn more at RocksDB.org.

For the purposes of today's tests, the CPUs tested based upon what I had available at the time were:

- 2 x Intel Xeon Platinum 8280 (Cascade Lake)
- 2 x AMD EPYC 7601 (Naples)
- 2 x AMD EPYC 7502 (Rome)
- 2 x AMD EPYC 7642 (Rome)
- 2 x AMD EPYC 7742 (Rome)

Thanks to Intel and AMD for providing the respective CPUs under test. Thanks go out to Gigabyte as well for the Gigabyte S451-3R0 server that continues to be used in our Cascade Lake testing. Each server was equipped with an Intel Optane 900p 280GB NVMe solid-state drive for storage and RAM for each CPUs maximum rated frequency and channels. Each server was running Ubuntu 19.10 with the Linux 5.3 kernel and EXT4 file-system.

Via the Phoronix Test Suite various operations were benchmarked with RocksDB in seeing how its performance compares as some complementary data to our recent PostgreSQL 12.0 testing and the several AMD EPYC Rome Linux reviews published thus far. Of course, I continue to run new workloads on these various servers especially as new test profiles are added to the Phoronix Test Suite / OpenBenchmarking.org. If there are any other test requests for workloads not covered yet (particularly for Phoronix Premium supporters with their higher priority feedback channel), be sure to let me know. Those wanting to see how their own Linux system(s) perform for RocksDB can repeat these tests by simply running phoronix-test-suite benchmark rocksdb or if wanting to do a side-by-side comparison to the results shown in this article can run phoronix-test-suite benchmark 1910173-HU-FBROCKSDB20.