AWS HA - instance sizing

Ohai!

I’d like to hear people’s experiences with the Enterprise Chef AWS HA
implementation https://docs.chef.io/install_server_ha_aws.html. In
particular, I’d like to know how folks are sizing the instances, and how
many nodes they’re handling. I’m looking to support around 2500 nodes for
the get go, but will need to support up to ~10,000.

Thanks in advance!

Hi Gabriel,

In my Chef server load-testing experiments I settled on an optimum size of
c3.xlarge for the frontends and c3.2xlarge for the backends. It’s
important not to skimp on backend storage - I used EBS-Optimized instances
and SSD-backed EBS volumes.

At 2500 nodes you could probably start with smaller instances, but it’s
highly dependent on your converge interval/splay and also the amount of
work each run puts on the depsolver and search endpoint.

*Irving Popovetsky *— Team Lead, Customer Engineering | CHEF

mobile: 503-490-5523 – irving@chef.io – *my: *Blog
http://irvingpop.github.io/ Github http://github.com/irvingpop Twitter
http://www.twitter.com/irvingpop Calendar Free/Busy
https://www.google.com/calendar/embed?src=irving%40chef.io&ctz=America/Los_Angeles&mode=WEEK

On Fri, Jun 12, 2015 at 8:02 AM, Gabriel Rosendorf grosendorf@gmail.com
wrote:

Ohai!

I’d like to hear people’s experiences with the Enterprise Chef AWS HA
implementation https://docs.chef.io/install_server_ha_aws.html. In
particular, I’d like to know how folks are sizing the instances, and how
many nodes they’re handling. I’m looking to support around 2500 nodes for
the get go, but will need to support up to ~10,000.

Thanks in advance!

Thanks, Irving. That’s extremely helpful.

On Jun 12, 2015, at 7:16 PM, Irving Popovetsky irving@chef.io wrote:

Hi Gabriel,

In my Chef server load-testing experiments I settled on an optimum size of c3.xlarge for the frontends and c3.2xlarge for the backends. It’s important not to skimp on backend storage - I used EBS-Optimized instances and SSD-backed EBS volumes.

At 2500 nodes you could probably start with smaller instances, but it’s highly dependent on your converge interval/splay and also the amount of work each run puts on the depsolver and search endpoint.


Irving Popovetsky — Team Lead, Customer Engineering | CHEF
mobile: 503-490-5523 <> – irving@chef.io mailto:irving@chef.io – my: Blog http://irvingpop.github.io/ Github http://github.com/irvingpop Twitter http://www.twitter.com/irvingpop Calendar Free/Busy https://www.google.com/calendar/embed?src=irving%40chef.io&ctz=America/Los_Angeles&mode=WEEK
On Fri, Jun 12, 2015 at 8:02 AM, Gabriel Rosendorf <grosendorf@gmail.com mailto:grosendorf@gmail.com> wrote:
Ohai!

I’d like to hear people’s experiences with the Enterprise Chef AWS HA implementation https://docs.chef.io/install_server_ha_aws.html. In particular, I’d like to know how folks are sizing the instances, and how many nodes they’re handling. I’m looking to support around 2500 nodes for the get go, but will need to support up to ~10,000.

Thanks in advance!