Advanced reconstruction


Advanced Reconstruction Updated Overview

Reading large data blocks from RAID storage appliances, normally requires that disk array reads from each drive, combines fragments and passes on to the external interface such as FC, iSCSI, IB or SAS. Each drive in the array has a difference speed, and even same model disks have slightly difference response times. Plus, seek time might differ depending on where the heads were on each disk in the array before the read request came in. Therefore, drives will complete the read request at different points in time.

With the standard, straight-forward approach, the system has to wait until the last drive completed reading the data.

How We Improved RAID 6 Performance

In the event of a disk failure, some of the information is missing and RAID uses special parity checksums, called RAID syndromes, to recreate the missing information.

RAIDIX developed a unique, patented RAID calculation algorithm using mathematical analysis of finite-number fields, ideally suited for parallelization. It can take advantage of long processor registers such as SSE-family available in Intel CPUs. This algorithm allows RAIDIX SAN appliances to restore the data, when multiple drives fail, as fast the disk controllers can read the data from the healthy drives.

The same algorithm is used to reconstruct the information, even in the healthy RAID array. This means that unlike competitors where SAN appliances have to wait for the slowest disk, RAIDIX only needs to read the information from the fastest drives. In RAID 6 configuration, we don’t wait for the last two disks. This allows RAIDIX to significantly outperform competing products. Performance benchmarking showed that on streaming workloads, such as those common in media broadcast and production, backups and HPC; Advanced Reconstruction alone yields 10% and higher speed improvement.