Re: Reliability models

[Loic Dachary <loic@dachary.org>]

[-- Attachment #1: Type: text/plain, Size: 1301 bytes --]

Crystal clear, thanks !

On 13/01/2014 18:25, Kyle Bader wrote:
> Hi Loic,
> 
>> IIRC you figured out how to use https://github.com/ceph/ceph-tools/tree/master/models . Do you happen to have some kind of HOWTO or even the list of commands you've used to get me started ?
> 
> git clone ...ceph-tools
> cd ceph-tools/models/reliability
> ./main.py -g
> 
> At this point a graphical form should pop up and most of it should be
> straight forward. NRE rate is non recoverable read errors, the NRE
> model determines how an array/disk responds to a non recoverable read
> error event. I tend to be on the conservative end having seen my share
> of RAID horror shows so I set the NRE model to "fail". The other
> gotcha is most disk manufacturers use MTBF or AFR while the
> reliability modeling tool requests FITS. The formula you will need is:
> 
> MTBF = 1,000,000,000 x 1/FIT. [1]
> 
> Stripe length is the number of RADOS objects that are require to store
> a blob of data, ie all RADOS objects that compose a RBD volume. Since
> this code was written a while ago there is no code to support modeling
> erasure coded objects. If you have any questions, let me know!
> 
> [1] http://en.wikipedia.org/wiki/Failure_rate#Units
> 

-- 
Loïc Dachary, Artisan Logiciel Libre


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 263 bytes --]