Comments on: MongoDB on ARM Processors

By: AC

AC — Wed, 27 May 2020 15:58:26 +0000

Can you rerun these tests comparing the new Graviton2 instances? It would be interesting to see how this compares

By: David Murphy

David Murphy — Thu, 18 Apr 2019 14:34:40 +0000

Did you happen to connect this to PMM? I am curious as to the CPU scaling perspective, after say 12 cores was x86 or ARM more effictive?

By: Adamo Tonete

Adamo Tonete — Mon, 25 Mar 2019 19:00:50 +0000

Hi HarpPDX,
We used m4.2xlarge and c5.4xlarge, I will fix the graphs. Thanks for letting us know.
Regarding the threads with about 150 threads we hit the arm limitation and the performance didn’t increased. CPU usage went to 100% and context switch killed the performance.
We will do a follow up article with more data soon.

By: HarpPDX

HarpPDX — Mon, 25 Mar 2019 18:08:24 +0000

Thanks for the well-written article…a couple of comments/questions:
– the result charts refer to different instance types than the table in the beginning of the article, can you confirm which instance types were used? i.e., c5.2xlarge or c5.4xlarge and, m4.2xlarge or m4.xlarge
– was there a reason you stopped at 128 ycsb threads when the performance was still increasing (compared to previous run at 64 threads)?
– I like your method of presenting normalized performance per $

By: Kay Agahd

Kay Agahd — Mon, 25 Mar 2019 08:48:01 +0000

Very interesting and well written. Thank you for the insights!

By: mdcallagk Callaghan

mdcallagk Callaghan — Fri, 22 Mar 2019 12:58:41 +0000

Thank you for sharing details that allow us to reproduce this. The open question is how large the price/perf difference needs to be to motivate a migration to ARM.