Date
1 - 5 of 5
Is Garden limiting disk IOs ?
|
Out of curiosity, I'd like to understand whether Garden is limiting disk
IOs, similar to [2]. I do see disk space, inodes..., and network IOs limits described into [1] but did not find block IOs. Thanks in advance, Guillaume. [1] http://godoc.org/github.com/cloudfoundry-incubator/garden [2] https://github.com/appc/spec/blob/master/spec/ace.md#resourceblock-bandwidth |
|
|
|
Will Pragnell <wpragnell@...>
Hi Guillaume,
toggle quoted message
Show quoted text
No, Garden-Linux does not currently do any disk IO limiting. Thanks, Will On 30 July 2015 at 17:24, Guillaume Berche <bercheg(a)gmail.com> wrote:
Out of curiosity, I'd like to understand whether Garden is limiting disk |
|
|
|
Thanks Will.
toggle quoted message
Show quoted text
Would other existing limits (e.g. nice, or fair cpu share) prevent one app from using up all of the host disk io, or could such app impact its neighbor containers (and other DEA/Cell process), and possibly up to denial of service (e.g. a compromised app that was to impact the cf platform that host them) ? Guillaume. On Thu, Jul 30, 2015 at 6:34 PM, Will Pragnell <wpragnell(a)pivotal.io> wrote:
Hi Guillaume, |
|
|
|
Will Pragnell <wpragnell@...>
The honest answer there is that I'm not sure. It's conceivable that
toggle quoted message
Show quoted text
limiting CPU for a container could prevent it from pummeling disk enough to DOS a garden host, but we've not tested that. I'll talk to the team about investigating this DOS vector. On 30 July 2015 at 21:03, Guillaume Berche <bercheg(a)gmail.com> wrote:
Thanks Will. |
|
|
|
Thanks Will. I guess with Garden moving to runC, the migration will bring
toggle quoted message
Show quoted text
disk io limits, would be nice to record that somewhere in a story so that this does not get forgotten. I was about to start some quick tests with a disk benchmark tool such as boonie [1], (invoked by the binary buildpack with a custom start cmd using netcat to bing the $PORT, invoke bonnie and return results in the HTTP response), and see how much other containers running the same test would be impacted. A refinement would be to have a regular app (e.g. a ruby app), and mesure response time for startup and response, with and without having one neighbor running bonnie on the same host. [1] http://www.textuality.com/bonnie/intro.html Guillaume. On Mon, Aug 3, 2015 at 11:10 AM, Will Pragnell <wpragnell(a)pivotal.io> wrote:
The honest answer there is that I'm not sure. It's conceivable that |
|
|