SAS Low-latency Backend
While clustered arrays use Gigabit Ethernet for their storage expansion fabric, Nimbus H-class systems use SAS. Not only does SAS provide 12 times greater bandwidth than Gigabit Ethernet, SAS also offers 85% lower latency, dramatically accelerating IO performance. Storage traffic can also be load-balanced across all ports using built-in multipathing support.
SAS v GbE in Bandwidth and Latency
Large Cache Is Essential
As storage is consolidated, especially in virtualization environments, IO patterns become increasingly random in nature. With multiple VMs sharing centralized storage, and each VM making its own IO requests, random IO grows even more prevalent. As a result, cache plays a vital role. Most IP storage arrays have 2 GB or less of cache, which is quickly filled up when there are terabytes of data on the system. Nimbus H-class systems have up to 24 GB of cache, providing a substantial buffer to increase cache hits and raise IO performance dramatically.
Policy-based Caching to Increase Hits for Top Applications
Most storage arrays cache data on the system fairly indiscriminately. With storage capacity often 1,000 times larger than the cache, ensuring cache hits for the applications that need it most is virtually impossible with traditional arrays. There needs to be a way to prioritize cache allocation for specific applications or virtual machines. Nimbus H-class systems feature an innovative caching algorithm that enables just that: IT managers can dynamically allocate the cache on a volume basis. This unique approach gives cache priority to the virtual machines and applications that need it most.
SSD Flash Technology
In addition to cache, SSD's (solid state drives) augment IO performance by eliminating the head seeks and rotational latency of hard drives with flash-based electronics. SSD technology ranges from consumer-grade MLC (multi-layer cell) flash to enterprise-grade SLC (single-layer cell) flash, which are ten times more reliable. These two flash types differ in their write life, or number of writes each bit of flash memory can sustain before the flash itself becomes unable to retain a charge, resulting in the equivalent of a "bad block". Nimbus only offers SLC-based SSD drives combined with controller-based wear-leveling, which spreads writes evenly across the flash to maximize its life.
Unparalleled IOps per Watt
The IOps performance of SSD's is staggering compared to hard drives as shown below. Even more compelling, SSD's offer 475x greater IOps per watt than 15K SAS hard drives, delivering the IO performance demanded by transaction-intensive applications in a fraction of the space and power.
SSD v. HDD Performance and Power
| |
SATA 7,200 rpm |
SAS 10K rpm |
SAS 15K rpm |
SSD |
Sequential
(MBps)
|
60 MBps |
95 MBps |
120 MBps |
150 MBps |
Random
(raw
IOps)
|
75 IOps |
130 IOps |
145 IOps |
8,100 IOps |
Power
(Watts) |
12 W |
14 W |
17 W |
2 W |
| IOps/W |
6.25 |
9.3 |
8.5 |
4,050 |
Multi-core, Scalable Internal Design
Nimbus H-class systems feature tremendous processing power. The systems are powered by eight 64-bit processor cores, providing 9 times the CPU horsepower of clustered systems. In addition, dual packet processing engines (PPEs) offload TCP processing from the main processor cores onto independent parallel-processing state machines. The PCIe Gen2-based internal architecture supports up to 140 Gbps of internal bandwidth between 10 GbE controllers, SAS expansion links, and the cache subsystem, nearly 10 times greater than competing designs. With such a powerful foundation, H-class storage systems can be expanded in capacity without adding additional processors, reducing storage cost and power consumption over time. |
|
| Key Features |
 |
| • SAS offers 12x more bandwidth than GbE |
| • Latency of SAS is 85% lower than GbE |
| • Increasingly random IO needs cache, SSD |
| • SSD IOps 50x faster than 15K SAS drives |
| • 9 times greater processing power |
| • 10 times greater system bandwidth |
|
|
|
|
