Cloud & Cluster Acceleration
Businesses are increasingly turning to general-purpose cluster, grid, or cloud infrastructures to support computing requirements. Whether private or public, the infrastructure must manage workloads to maximize customer benefits, meet SLAs, and lower operating costs.
The flexibility, rapid reconfigurability, and high utilization rates of virtual and cloud architectures have sped their adoption in many data centers. Meanwhile, the economic benefits of commodity servers and the combined power of clustered systems drive their adoption for heavy enterprise and analytic workloads.
Cloud and clustered infrastructure can put extraordinary demands on storage that are not satisfied by typical enterprise storage architecture. Enterprise storage upgrades are expensive, they require additional space, power and cooling and they often add a layer of complexity to the architecture.
RNA MVX aggregates existing RAM resources into an extremely fast caching layer without requiring specialized hardware or changes to the existing storage approach. The result is a speed upgrade that is transparent to both applications and storage, and which does not expand the datacenter space, power or cooling footprint. In dynamic environments with mixed workloads, RNA MVX can adapt to the memory needs of various applications with rapid and flexible provisioning of virtual memory resources.
Scalability and utilization
Cloud and cluster architectures share the related concerns of scalability -- how to handle ever-larger loads -- and utilization -- how to use resources efficiently at any level of load. If a cluster or cloud architecture has declining utilization as it grows, it won’t scale well. Conversely, an architecture that improves its utilization as it grows will scale well.
MVX private cloud technology creates a large distributed memory service
- a Memory Cloud - from RAM in existing servers
MVX solves the utilization and scalability problems common in clusters
and clouds
It is typical that storage bottlenecks and RAM limitations arise as load grows. With increasing numbers of virtual machines in a cloud environment, or increasingly large data sets in a cluster setting, the storage demands approach the physical capacity of storage equipment, and I/O waits become longer and more prevalent. This causes delays throughout the cluster or cloud, as processes are blocked and RAM and compute resources are tied up waiting for I/O to complete. Multicore processors offer ever-increasing compute capacity, but I/O wait and RAM contention limit their ability to perform.
RNA MVX directly eases these limitations by creating a Memory Cloud to apply RAM where it’s needed. For example, Memory Cache and Memory Store capabilities keep data in the compute tier to cut I/O wait times. When the same data is needed in multiple nodes, as in Hadoop’s map/reduce or other distributed functions, the effect is dramatic. At the same time, Memory Motion leverages RAM as swap space, so that analytic jobs in clusters and hypervisors in clouds can smoothly expand their addressable memory space to tackle larger problems and service more virtual machines.
The result can be measured in elevated CPU utilization -- 4X better utilization in some cases. And since MVX employs more RAM and more network interfaces every time a server is added, the improvements in utilization persist as demands and infrastructure grow. RNA MVX is a direct solution to the utilization and scalability problems that are most common in clusters and clouds.
MVX is the superior alternative
MVX is unique in being software-only, delivering microsecond access rates while utilizing existing hardware without application tuning, disruptive server upgrades, or additional layers of infrastructure.
MVX provides resoruces with the speed of DRAM and the capacity of traditional storage solutions
Other alternatives fall short of what MVX provides. SMP or global shared memory systems are costly, have scaling restrictions, and involve additional hardware that’s used for only specialized analytical workloads. Configuring ‘fat’ nodes with large amounts of RAM is far more costly than using commodity servers, and they are less flexible for cloud or changing cluster needs. Fast conventional storage is 100X slower than memory virtualization using MVX, while SSD is still 10X slower than MVX and is not a shared resource.
MVX creates multi-terabyte memory clouds of sharable RAM from existing cloud or cluster resources, reducing data replication and data management overhead while increasing overall utilization. The result is a more scalable and fundamentally more cost-effective architecture for most forms of distributed computing.
