Scality adds all-flash support to RING object and file storage

Scality has added all-flash support for RING8 object storage software. The tweak takes Scality RING8 into new performance-sensitive file and object  workloads.

Scality RING already supports flash use for metadata and internal indexes but data payloads were stored on hard disk drives until now. The company thinks it is now practical and cost effective to support data payloads on flash, thanks to the decreasing cost of SSDs, through TLC and QLC 3D NAND.

Giorgio Regni, Scality CTO and co-founder, issued a quote: “This isn’t Scality’s first rodeo with flash. We have a decade of experience using flash media in the RING architecture to accelerate a high percentage of operations for file and object access.”

Giorgio Regni.

These workloads include single-tier CCTV, fast tier1 backup and restore, tier 1 medical imaging (PACS), and Big data analytics on file. Scality also has a scale-out file storage capability and notes many analytics applications such as Splunk support file storage while also embracing the S3 object API.

The company claims object storage competitors that support all-flash typically do not have an integrated file access capability. They rely instead on a file access gateway which can limit scalability and performance to the file end-point servers. Scality claims RING8 performs faster than these competitors, without supplying examples, and scales out. It says vendors with integrated file capability – such as Pure Storage’s FlashBlade – use proprietary hardware.

Benchmarks

In testing, a RING8 system had data payloads stored in six HPE Apollo 4200 servers, each with 16 x 12Gbits/s SAS SSDs, 96 SSDs. This setup delivered more than 10GB/sec to accessing clients across a 40Gbit/s Ethernet link, fully saturating it.

The six Apollo boxes were connected across 100GbitE to a seventh Apollo Connector server running S3/SMB software, and the >10GB/sec was demonstrated with S3 and SMB.

Five client systems, with 1MB block sizes, read and wrote 1GB objects or files from the Connector. RING8 used 4+2 erasure coding, so 50 per cent more data was written to RING8 from the Connector.

A Scality FAQ claims “at least a 50 per cent performance boost from using just SAS HDD (since we are network limited, it is possible the increase is even higher.)” The company notes that the SMB file system protocol lack parallelism and so there are welcome performance benefits from scale-out RING8’s SSDs. 

Scality reckons RING8 all-flash performance will increase as more nodes are added and the company will run tests with 100GbitE client access networking to check this out.

It said lower latency makes all-flash RING8 especially suitable to applications needing random access to small files. There is proportionally less time saving with large files. It also says many if not most object workloads today feature large objects. Therefore all-flash RING8 can serve both needs whereas the random-access small file workloads were not previously available to it.

Cloudian added all-flash support to its HyperStore object storage last week, saying it can deliver over 2GB/sec read performance per node. Six such nodes would deliver 12GB/sec read performance.