A start-up has found a way to aggregate memory and share it across servers, improving performance of online transaction processing and clustered or grid computing.
RNA said the virtualisation of memory removes latency penalties that affect application performance, and that its technology is already being used by a multi-billion dollar global hedge fund to speed up the processing of trades. RNA also has a partnership with Colfax International to combine its memory virtualization software with Colfax's high-performance computing systems.
RNA CEO Clive Cook, previously CEO of wireless broadband company Verilan and telecom software vendor Elematics, calls memory "the third wave of virtualisation," after servers and storage.
"Memory virtualisation, like server and storage virtualization, offers the benefits of consolidation and compelling cost savings," Cook writes in an article on VMblog. "Yet over and above early server and storage implementations, in 2009 memory virtualisation will introduce a new way of thinking about virtualisation that challenges the way IT manages applications, clustering, the data centre, services and business itself."
While the words memory and storage are often used interchangeably, Cook says they are not synonymous, noting that data is retrieved from disk storage and then put into memory before it can be used by the processor. Adding storage doesn't necessarily solve problems related to application performance, he writes.
Virtualising memory extends the memory available to a server beyond its own physical capacity, allows shared memory in clustered or grid computing environments and could be a key enabler of cloud computing in enterprise data centers, he wrote.
The speed of an application is often limited because its working data set exceeds availably memory in the physical server, but with memory virtualisation the entire working data set can be loaded into memory allowing the processor to access it directly, he explained. While VMware's hypervisor allows the sharing of memory across virtual machines, Cook sees RNA's technology as being complementary to server virtualisation rather than competitive to it.
"Memory virtualisation paired with x86 server virtualization will drive the adoption of virtual machines into the production data centre," Cook argues. "Today, server virtualisation has been limited to operational applications, not business-critical applications due to performance and reliability concerns."
Further, virtualisation lets memory be distributed and shared across the network, making multiple terabytes of memory available to any application.
RNA lists two products on its website: RNAmessenger, designed for trade executions with low latency and high throughput; and RNAcache, which creates a shared, network-attached cache of virtualised memory.
Two investment banks which are using the latter product sped up read times by a factor of 18 and write times by a factor of 35, RNA said on its product site.
"RNAcache caches both data and metadata across nodes linked by common or high-speed fabric, GbE and 10GbE/InfiniBand," the company stated.