S Class‎ > ‎



Certainly one of the top benefits of an all-flash storage system is exceptional performance. However, not all flash storage systems are created equally. With the NAND silicon capable of over 100x faster response time than hard drives, it is vital to engineer the storage system specifically for flash to capitalize on these gains. Simply retrofitting SSD’s into an existing storage array is no different than strapping a jet engine to the back of economy sedan. Don’t expect proper performance nor efficiency.

Performance starts at the NAND silicon itself, so let's analyze performance from the component upwards. Each Nimbus Enterprise Flash Module (there are 24 per Nimbus S-Class enclosure) features sixteen NAND chips managed by a dedicated flash processor. This flash processor fully offloads all flash management from the S-Class system processors. Since flash management is a very specialized function, involving millions of ECC calculations, garbage collection, read disturb management, and much more, offload is critical to avoid distracting the system’s valuable CPU resources.

In addition to managing flash, each flash processor also has non-volatile write cache. This write cache helps accelerate performance in write-intensive applications by considering IO’s committed once they arrive in the write cache, rather than when they are committed to the NAND silicon itself. Since flash programming takes longer than DRAM programming, this approach improves performance without having to rely on any centralized cache and associated cache coherency architectures. The cache within the flash processor is kept alive via capacitors in the flash module itself, so no battery backup is needed to preserve data integrity. Cache destaging is automatically managed by the flash module independent of the system itself.

As stated, there are 24 flash modules in a Nimbus S-Class system. How each module connects internally to the Nimbus S-Class system is essential, and in this case, a native 6 Gbps interface is utilized. The 6 Gbps interface provides up to 600 MBps of performance, enabling the flash module to achieve its full native performance without any bottlenecks. The interface also supports hot-addition and hot-removal of flash modules without any system interruption, facilitating online expansion and in-production servicing.

Since flash is so fast, just a few flash modules would saturate a conventional expander-based storage enclosure that relies heavily on oversubscription. When designing the S-Class, Nimbus therefore implemented a new “non-blocking” midplane that provides full line-rate bandwidth to every slot in the enclosure simultaneously. Each of the 24 slots has a dedicated 6 Gbps of bandwidth to the S-Class CPUs, a total of 144 Gbps, which is 6 – 12 x more than conventional arrays. As  result, every flash module can harness full bandwidth simultaneously, delivering exceptional throughput while minimizing latency.

The CPU’s come next. The Nimbus S-Class is powered by 12 physical Westmere cores, offering 24 virtual cores with Hyper-Threading. Along with DDR3 DRAM, there is ample processor horsepower to deliver up to 800,000 IOps and 8,000 GBps of performance from one S-Class system. In fact, the CPU’s are only modestly utilized even when delivering this performance. This is to ensure that there are ample CPU resources for software services provided by the HALO software, such as deduplication, thin provisioning, and much more. It’s interesting to note that one reason these CPUs run so efficiently is because they are communicating to flash memory. CPU’s burn a lot of cycles needlessly in “IO waits”, a vicious process whereby the CPU is “waiting” for data from hard drives before being able to move to the next operation. Unfortunately, IO waits tie down the CPU from doing other things, reducing overall performance. Thanks to the Nimbus’ 100% flash architecture, IO waits are minimized, keeping the CPU’s humming along very efficiently.

If you have come this far, you want to see how this story ends. It would be unfortunate to let all this performance go to waste because of inferior network connectivity. That’s why the Nimbus S-Class provides huge amounts of bandwidth via four IO modules, each supporting up to 40 Gbps of bandwidth. That’s 160 Gbps of network-facing bandwidth, a near perfect match for the 144 Gbps of backend bandwidth, ensuring again that there is no funneling of IO. The datapath remains 100% free of oversubscription. Customers have their choice of 10 Gigabit Ethernet, Fibre Channel, Infiniband, and even Gigabit Ethernet IO modules for these four slots. Depending on your network environment, simply select the proper IO modules for you. The S-Class fully supports multipathing IO across all ports in parallel, enabling up to sixteen 10 Gigabit Ethernet ports, eight QDR Infiniband ports, or eight 8 Gbps Fibre Channel ports, or any mixture of them.


Virtual Data Products, LLC - P.O. Box 492575,  Lawrenceville, GA 30049 ph 770-995-7298

A Proud American Company !