Key customer benefits
Tightens read tail latency
Scales RAID performance for NVMe SSDs
Slashes Cost per TB of data stored
There are two key challenges faced by modern RAID solutions.
- SSD IO performance is scaling rapidly, with individual NVMe SSDs delivering millions of operations per second (IOPs) and Gigabytes per second of throughput.
- Latency, specifically read tail latency, has become the key performance metric in the datacenter.
The scaling of SSD IO performance creates an increasing potential for bottlenecks in the traditional IO flow and RAID architectures. This can cause spikes in read tail latency and prevent users from getting the full performance from their NVMe SSDs.
The SSDs themselves contribute to the tail latency challenges. As users push their SSDs with intensive mixed read-write workloads, latency variation climbs rapidly. To meet latency SLAs, users have to operate their drives at levels far below their maximum performance. This results in poor scaling of performance and increases the hardware (and power!) needed.
The combination of SupremeRAID™ SR- 1000 with the ScaleFlux CSD 3000 offers a solution to the RAID performance scaling and quality-of-service (QoS) challenges.
SupremeRAID adopts a heterogeneous architecture where compute-intensive RAID calculations are offloaded to a GPU. The massively parallel vector processing capability of a GPU aligns well with calculating RAID parity data for many simultaneous IO operations.
The ScaleFlux CSD 3000 integrates hardware-accelerated data compression and decompression, resulting in consistently high Read-Write operations, consistently low latency, and lower cost per GB of data stored. This allows users to push each SSD to higher levels of performance without sacrificing latency consistency. Also, with the compression feature, the CSD 3000 can store more data per GB of Flash. In the white paper, the CSD was formatted to 2x its physical capacity, reducing the costs of data storage without sacrificing performance.
The results show that the SupremeRAID array with ScaleFlux CSD 3000 NVMe SSDs enabled all the resources of each SSD to be leveraged by a single logical volume.
The ability to pool NVMe SSDs and provision logical volumes that can utilize the full parallelism provided by the underlying storage can be used as a critical tool to manage tail latency. In combination with the reduced Read-Write interference inherent to the CSD, this combination can be used to:
- Create virtual volumes that outperform any physical volume and provide consistent, low tail latency at multi-million IOPS performance levels.
- Avoid stranding SSD performance by many virtual volumes to use the full performance provided by the drive pool. This is particularly suited to bursty workloads where a single virtual drive can deliver many millions of IOPS and low latency at any capacity point.
- Utilize the CSD’s compression to double its effective storage capacity – slashing $/TBe (cost per Terabyte-effective).
Graid Technology + ScaleFlux
Improve User Experience & SLAs by using Accelerated RAID and Hardware-Compression to control tail latency.