Storage Offload

Storage Offload with MVX Memory Cache

As applications grow in data centers, whether by adding users or through increasing sophistication, the storage systems in the data center feel the load.  The scalability  of storage performance is an ongoing and recurrent problem in most data centers.  Many new types of applications and architectures make intensive demands upon fast network storage, including virtualization, clustering, and cloud computing algorithms like Hadoop.

To support more storage traffic, data center managers currently plan and commit to storage system upgrades.  These upgrades are usually complex and expensive.  Technologies that offload storage demand without replacing hardware can cut both capital costs and operational complexity, extending the useful life of existing equipment.

RNA Networks' MVX product supports storage offload with centrally managed facilities to reduce storage load.  MVX creates a Memory Cloud by aggregating RAM from servers around the data center.  This pool of RAM can be used as a Memory Cache which allows storage clients to satisfy most of their NAS storage requests in the compute tier without burdening NAS systems.  Any number of Memory Caches can be configured from the Memory Cloud.  At the same time, the Memory Cloud can be deployed as “networked RAMdisks” using Memory Motion and Memory Store, to keep swap or temporary storage space in RAM and thereby remove that load from disk-based storage systems.

When software algorithms require multiple reads of the same data sets, the resulting performance improvements can be dramatic.  For Hadoop-style map/reduce functions on large data sets, NFS reads can be cut by 99%; parallel reads are satisfied by a single copy of the data set spread across the RNA Memory Cloud.  With data delivered by multiple MVX nodes across the data center, both network and storage system bottlenecks are eliminated.

MVX provides additional performance benefits beyond the storage system offload.  When providing storage services through the RNA Memory Cloud, I/O wait times are cut dramatically.  This can result in higher CPU utilization and reduced execution time for many processes.  It can also speed up every task in affected servers by rapidly freeing server memory resources that are dedicated to I/O.  In effect, all server tasks run better when I/O wait is reduced.  Application CPU utilization increases of 3x have been measured due to reduced I/O wait times.

The flexibility provided by MVX is particularly valuable for addressing storage offload.  The RNA Memory Cloud is normally configured as a single resource, aggregating RAM from servers anywhere in a data center into a single pool.  This pool can be quite large, reaching well over 1 TB if desired.  Large segments of the Memory Cloud can be concentrated in cache and RAMdisks to offload the particular storage systems that are most heavily loaded.  The size of the Memory Cloud provides RAM resources that can be far larger than the physical capacity of any any single storage system or server.  MVX provides a uniquely powerful technology for improving storage performance.