Buying the right storage is critical to making your cloud effective, but identifying the right solution is difficult. Drawing on 8 years of experience as a storage developer, this talk will help you learn the truth about how systems work. Topics range from the obvious-but-frequently-misplanned IOP/bandwidth/capacity needs of your system, to geeky-but-critical design choices (Consistent or Available under failures?), and on to what kinds of questions will let you evaluate and cost specific features. With examples in block devices and object storage, these are general questions and not about any particular solution; if you follow through on these you will be able to identify significant tradeoffs and weaknesses anywhere. If you’re designing a new cloud, realize you need to re-do storage in your cloud, or are just curious about how storage developers evaluate storage systems, this talk is for you!
This talk is part of my new self-appointed mission to produce educated storage consumers (I spoke at Boston’s Open Source Days on the costs of snapshots in Ceph). It is *not* a sales pitch and I expect not to refer to any products or non-OpenStack projects by name. I plan to walk through specific features of any storage system that administrators should understand but which vendors often obscure or try to distract buyers from. In particular, I’ll discuss how system design choices actually impact cloud admins/users beyond the feature list checkbox: ways in which failures can manifest and how they are contained (or not!) by the system, how to evaluate add-on features such as geo-replication or snapshotting, etc. I’ll also discuss how to usefully evaluate vendor benchmarks and comparisons.