Last year, Dell discreetly bought RNA Networks, an ISV mostly known for its software clustering solution that allows a group of servers to share their memory space to form a coherent shared memory pool. During a keynote at its 2012 Storage Forum, the firm has unveiled what it calls “Project Hermes”, whose goal is to apply this technology to the (radical)acceleration of Storage I/Os.
RNA’s Virtualization technology allows the members of a cluster to agregate their memory memory into a single shared pool – that memory can be used by apps and also used as a virtual storage block device, very similar to a RAMDisk. What is less known, is the fact that the same technology can also be used to pool the Flash capacity of the individual x86 nodes into a single shared storage space. Under “Project Hermes”, Dell intends to make use of these two capabilities to dramatically boost the I/O performance of a cluster of servers connected to a SAN. The idea is to use the technology to provide a layer of read/write memory cache (literally Tier minus 1) and read/write Flash cache (Tier 0)– using the PCI-express SSDs that’s plugs into Poweredge 12G servers. These two cache layers would then be used to provide a massive level of acceleration of I/O operations.
When Dell acquired RNA Networks the technology was able to aggregate up to 1024 servers cores into a single RNA Networks (it seems Dell has substantially improved that to scale to “hundred of servers” according to a company source. The fastest Dell PowerEdge 12G servers are able to accommodate up to 1.5 TB of memory and have four slots for Micron SLC flash drives directly connected to the PCI-Express Bus.
At least 32 of these servers – with the 1024 core limitation -could be aggregated into a single cluster by RNA Networks technology, enough to assemble a system that can address 48 TB of RAM and about 75 TB of Flash. That capacity of RAM could be made available to in-memory applications (eg a database), but also to a cache fronting the SAN arrays. By the way RNA’s technology ensures cache coherency between the cluster nodes and provides redundancy to protect the cluster against the failure of multiple nodes. To make it short, RNA Networks magical sauce allows Dell to solve a set of problems that all of its competitors, starting with EMC, IBM and HP are still fighting with(most of them provide flash in servers, for storage or cache, but haven’t figured out –yet- the way to share them across multiple nodes, which for exemple limits their caching feature to read cache).
The most amazing thing is probably the versatility of the RNA Networks’ technology. It can be applied to the acceleration of I / Os other a network, but is also suitable for applications requiring large amounts of memory (or shared memory space between nodes). It could for example be used to create systems for analytics – to compete with systems like oracle Exadata or Exalytics-, to host large scale “in-memory” databases (why not a Dell/RNA cluster running SAP HANA), or to run HPC apps requiring large scale memory space.
Note : we’ve been told that RNA Networks’ technology could eventually be embedded under the array OS to provide a large scale clustering capability for controler nodes. That would require moving certain features (like snapshotting) from the array software to the RNA layer. Today it is still science Fiction but we could imagine a large scale cluster of compellent controller nodes in front of a massive amount of storage. Like they say in Texas that would be “bigger’n Dallas”… and would probably scare the hell out of EMC, Hitachi, NetApp and HP