Cloud Storage Performance

 

Zetta Storage Service is designed from the ground up to deliver scalable, guaranteed performance in a multi-tenant environment. Aspects of cloud service performace include the backed system itself, the communication to the service and performance gurantees.

 

Back-end System Performance

Most storage virtualization architectures take a monolithic storage array and sub-compartmentalize it into smaller logical partitions to use it for multiple applications simultaneously.

 

The Zetta cloud storage solution is effectively the inverse of this. Zetta aggregates a myriad of smaller nodes and logically combines them into a very large virtual array, enabling far more dynamic scale-out capabilities for space and performance, and greatly increasing the fault-tolerance of the overall system.

 

The Zetta storage system is a three-tiered architecture with multiple servers at each tier for both performance and resiliency. The tiers communicate with each other over multiple scalable and redundant carrier-class 10 gigabit and 1 gigabit switching networks.

 

  • The front-end access tier connects to the outside world via multiple 10 gigabit circuits and is optimized for throughput.

  • The middle tier, the controller tier, performs metadata and RAID functions and is backed by multiple banks of redundant solid-state disk caches. This tier is optimized for IOPs.

  • The third tier, the storage tier, contains multiple servers with direct-connect SATA storage. The storage tier performs encryption, data storage and data validation and recovery and is optimized for storage density and performance.

 

Zetta leverages this entire scalable storage grid to host the Zetta File System. The Zetta File System is designed as a multi-tenant solution, leveraging virtualization at the customer access layer with per-volume encryption to ensure that all customer data is kept completely private and logically partitioned.

 

Network Performance

Zetta supports significant bandwidth with a 10 gigabit connection from the network coming into each node in our front-end access tier. Clients connect to Zetta in three general ways:

 

  • Over the Internet
  • Over a private connection
  • Over a direct datacenter connection

 

Performance in each case is different, but with today's bandwidth pricing and the appropriate use of protocol paralellization, it is possible to transfer a terabyte of data a day or more even over Internet connections and over a thousand miles distance. This far exceeds the requirements of all but the most data intensive applications. With direct connection and shorter distances, performance can approach on-premise networked storage performance.

 

“Zetta has been able to introduce the granular control necessary to guarantee performance and make it always available, tolerate multiple disk and system failures in any layer of their infrastructure, and turn data integrity from a passive “hoped for” characteristic to an active, guaranteeable process.”

— Jeff Boles, Taneja Group

Read the paper »

Performance Guarantees

Zetta provisions scalable virtual volumes and also guarantees the performance of each individual virtual volume, regardless of the number of other customers or workloads accessing the cloud storage service at the same time, by leveraging the very high degree of parallelism available in our distributed computing infrastructure. With strict service level agreements wrapped around availability, integrity and performance we are ensuring our customers receive the expected performance at ALL times.

 

For more information, read Chris Schin's blog entries including “Non-blocking performance.”