One of the more problematic challenges in the world of Big Data is how to move all that BD goodness around. Startup A3CUBE has a novel answer; the company has adapted the high-speed serial computer expansion bus standard PCI Express (PCIe) as an “intelligent network fabric.” The result:
A3CUBE’s In-Memory Network technology, for the first time, allows direct shared non-coherent global memory across the entire network, enabling global communication based on shared memory segments and direct load/store operations between the nodes. The result is the lowest possible latency, massive scalability and disruptive performance that is orders of magnitude beyond the capabilities of today’s high performance network technologies such as GigE, InfiniBand and Fibre Channel.
A3CUBE claims their RONNIEE Express data plane can provide internode latencies measurable in 100’s of nanoseconds and memory mapping that is 10 times faster than traditional PCIe memory mapping.
The company claims that a 5-node storage network using a multipoint direct topology with this RONNIEE Express products and 40 TB of low-cost SSDs can deliver a staggering 4 million IOPS. On the Sockperf benchmark RONNIE Express scored nearly 400,000 messages per second compared with the performance of 10 GBe cards from Chelsio and Intel which rated 100,000 and 50,000 messages per second respectively. This equates to something like eight times the performance of competing solutions for about one third of the cost.
The RONNIEE Express line currently consists of three products:
- The RONNIEE 2S, a compact PCIe-based network interface card
- the six port RONNIEE RIO (the multiple ports allow for high performance, minimal latency configurations such as toroidal and hypercubodial configurations)
- The RONNIEE 3 switch for building high-performance network fabric. The connectivity of these products is supported by A3CUBE’s ByOS a “scale-out operating system” that provides data protection without RAID overhead.
It’s all very impressive and a compelling pitch given the remarkably low cost and amazingly high performance claimed. My only criticism of the entire pitch is that they claim that they:
… spent more than 5 years looking at how the brain sends and receives signals. They did this brain mapping work to understand how to get past the limitations of today's networking technologies including Ethernet, InfiniBand, Fibre Channel and PCIe.
I see absolutely no justification for this claim. The physical interconnect architectures they suggest (toroids, hypercubes, etc.) have no relation to known human brain functioning or brain architectures and if their ByOS control system has any brain research derived attributes A3CUBE have completely failed to explain the relationship. I could be wrong but this sounds like marketing in overdrive and does nothing to support a product line that otherwise would appear to have tremendous potential.