SDC 2017 - Storage Performance and Reliability/Redundancy Evaluation - Junji Arakawa, Kei Kusunok

Downloand Presentation:

Abstract:
As one of the cloud providers, we have evaluated various kind of storage array for our IaaS infrastructure. Due to unexpected customer use-case on public IaaS, it’s difficult to model appropriate workloads for evaluation and to determine criteria for investment. Also, as you know, there are a number of virtual machines on a single storage array. These VMs constitute together modern/complicated ICT solution. Once even a storage component fails, a non-negligible impact will be spread on many customers.

Therefore, the providers may be tempted to have extra redundancy for their peace of mind, such as double parity, triple parity, more hot-spare, double backup etc. However such redundancy will cause several-fold expense. Even if we have such extra redundancy, there is usually a significant performance impact on the recovering process. So how to determine the reliability requirements/policies under cost limitation and evaluating performance degradation during failures are pivotal points on adopting new products.

In this session, we explore how to model IaaS workload and our practices to take how much redundancy on scale-up/out storage array and evaluate performance result from defined workload with normal/failure situation.