openstack team mailing list archive

[Nikolaus Rath <Nikolaus@xxxxxxxx>]

Hi Andi,

My perspective is that I'm working on an application that should work
with arbitrary service providers using swift. Therefore, I'm interested
in the minimal set of guarantees that I can always rely on, no matter
how the service provider has configured his particular swift instance.

Best,
Nikolaus


On 01/20/2012 04:10 PM, andi abes wrote:
> I'm finding this thread a bit confusing. You're comparing offered
> SERVICES to Software. While some of the details of the software will
> dictate what's possible, some are heavily dependent on how you deploy
> the swift software, and what kind of deployment decisions you (or your
> service provider) make.
> 
> As an extreme example - if you deploy 1 container server in a highly
> available fashion (hardware style), the you probably could get
> consistent container listings in the different update followed by read
> scenarios. Hosting huge swift installations with such a setup is not
> realistic - but that doesn't say you can't do that.
> 
> Similarly, swift offers quite a lot of flexibility in setting the
> eventual consistency window sizes (replication frequency, rates and
> such). So, while there are theoretical answers to missing replicas, the
> likelihood of those occurring depends on your deployment and operational
> practices employed. (e.g. how many replicas are made, how quickly are
> failed nodes/drives fixed and their content replicated to their
> replacement etc). 
> In the amazon case, much of this is captured in the 17 9's or the 3 9's
> guarantees for the reduced redundancy class. 
> 
> If your approach is from an API perspective, then issues around # of
> replicas (which is deployment parameter) are probably not relevant - if
> you trust your provider.
> 
> If your approach in this is from a Swift developer / deployer
> perspective  - then nvm. keep asking, cause it's much easier to read
> email than python ;)
> 


   -Nikolaus

-- 
 »Time flies like an arrow, fruit flies like a Banana.«

  PGP fingerprint: 5B93 61F8 4EA2 E279 ABF6  02CF A9AD B7F8 AE4E 425C