What do Virtualization and Cloud executives think about 2010?  Find out in this VMblog.com series exclusive.

Contributed article by Jim Bole, VP Products and Services, RNA Networks

Tapping the Potential of Memory Virtualization in 2010

In 2009, we saw a number of vendors turn their attention to addressing memory limitations in the data center.  From the Intel Nehalem processor's integrated memory controller and ability to address more DIMMS, to Cisco and their UCS Extended Memory Blades, it seems that the core data center infrastructure providers are working hard to innovate around server memory limitations. This type of validation for the problem has been welcomed and extended by vendors like RNA networks who launched similar software-based offerings earlier this year, with broad support for leveraging server memory footprints across multiple platforms and operating systems. All of these advancements point to the realization that by combining large memory servers with unused network resources (memory), companies can reduce IT infrastructure costs and complexity and extend the life of their infrastructure investments. 

Just as a desktop computer slows down when its memory gets full, limited memory prohibits the data center from operating at peak capacity.  This is a great use case for virtualization, and we expect the focus on "Memory Virtualization" to continue well into next year.  After all, who wouldn't want to improve performance without investing in expensive hardware upgrades?  On that note, RNA networks expects demand for memory virtualization to accelerate in 2010, fueled by the following predictions.

1. Storage will Run out of Headroom

Storage is often looked at as a way to address I/O bottlenecks and  memory capacity limitations, but this approach does not address the core I/O issue without large latency and throughput implications. Storage acceleration can only marginally improve application performance, whereas memory is 100 times faster than storage and is ‘application-aware'.  Meaning, memory operates at the application layer where it can more easily and quickly interact with active data - providing orders of magnitude increases in utilization, performance and speed. Regardless of the type of storage (spinning or SSD) or the acceleration options (NFS, storage caching), memory will remain the fastest option.

2. Data Volumes will Outpace Available Server Memory

The rate of proliferation and amount of data in the data center that must be quickly analyzed and interpreted is growing exponentially.  Memory density is not, and the density improvements available (16GB DIIMS for example) have cost implications that make them prohibitive for many applications. Methods to make data available to distributed servers are expensive in terms of programming complexity, bandwidth and IT oversight.  Not to mention the scaling issues.  As data volumes continue to grow it will outpace available memory on a server.

3. Memory as a Service in the Cloud

The success of a cloud, whether hosted or within the enterprise, hinges on its ability to ensure SLAs and keep costs down by harnessing as much utilization of the infrastructure as possible.  Tools to provision servers, storage, and I/O bandwidth are common in the cloud.  In today's world, memory is available as a component of a server - requiring it to be provisioned and managed in concert with processor capacity. But in practice, many application workloads are either highly compute bound (requiring minimal memory) or highly IO bound (requiring minimal CPU power but access to large data spaces). Yet these two resources are tightly coupled in terms of their provisioning and management, resulting in "lumpy" allocation of resources and suboptimal utilization. Memory virtualization breaks the fixed relationship between memory and servers, giving cloud providers far more flexibility and scale. Pools of networked memory are fundamental requirements for clusters, and will be a key component of the cloud infrastructure.

4. The Concept of ‘Caching 2.0' will Take Off

Changes in the data center landscape will give rise to a new caching model that reduces replication, enables large network-based in-memory application sets, and supports a broad range of data types and scalability.  Conventional alternatives such as adding more database nodes, expensive storage or faster processors can no longer keep data centers running effectively and efficiently.  Caching 2.0 dynamically scales as the volume of data in the enterprise increases. By applying memory virtualization across the existing infrastructure and enabling terabytes of active data to be kept in memory, companies will be well-equipped to stay ahead of the curve. We believe this memory-based application architecture will emerge over time as a foundational pattern for optimized data center operations.

About the Author

Jim Bole is vice president of products and services at RNA networks (http://www.rnanetworks.com/), a leading provider of memory virtualization software that transforms server memory into a shared network resource. Jim brings 25 years of enterprise experience in areas such as SOA, BPM and content management. He holds a Bachelor of Science in Applied Mathematics/Computer Science from Carnegie Mellon University and is a Board Member of DPAC Technologies (NASD:DPAC).