couchdb-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Stanley Iriele <siriele...@gmail.com>
Subject Re: Respawning server died, can't figure out why
Date Tue, 20 Aug 2013 17:11:49 GMT
If you're doomed to aws and being backed by ebs... Try going down the
provisioned iops... I saw significant performance gains when doing this...
That...and I picked a box with a good bit of RAM.

Couchdb makes excellent use of RAM! "Cache all the things!!"
On Aug 20, 2013 9:42 AM, "Joan Touzet" <wohali@apache.org> wrote:

> On Tue, Aug 20, 2013 at 09:48:59AM -0500, Nathan Vander Wilt wrote:
> > Thanks for these suggestions, Joan. Unfortunately it seems the server is
> under quite the opposite situation though: this is an m1.medium instance
>
> Smoking gun. AWS instances are notorious for bad disk throughput.
> Ultimately, CouchDB is heavily driven by disk performance.
>
> If your instance is EBS-backed, You're going to observe a ton of
> latency and worse; it's a bit of an open secret that Amazon's
> internal networking is oversubscribed, especially in us-east.
>
> You can prove it to yourself by swapping out this instance temporarily
> with a physical machine with approximately the same configuration.
> m1.medium is 1 CPU, 4GB of RAM, so your average laptop - try booting
> Linux from a thumb drive and running off of a dedicated HDD, even over
> USB. You'll find it performs significantly faster.
>
> If this works well, you can try running AWS instances from ephemeral
> storage instead of EBS for a bit of a boost, but you'll still be running
> into other unknowns on the backend. You may want to consider VMs from
> other companies in nearby data centers if you have your infrastructure
> in us-east-1, such as SoftLayer's Washington DC data center, NAC in NJ
> or Rackspace in Chicago.
>
> Good luck,
> Joan
>
> --
> Joan Touzet | joant@atypical.net | wohali everywhere else
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message