The DataVault was Thinking Machines' mass storage system, storing 5 GB of data, expandable to 10 GB with transfer rates of 40 MB/s. Eight DataVaults could be operated in parallel for a combined data transfer rate of 320 MB/s for up to 80 GB of data. [1]
Each DataVault unit stored its data in an array of 39 individual disk drives with data spread across the drives. Each 64-bit data chunk received from the I/O bus was split into two 32-bit words. Each word was combined with 7 parity bits to form a 39-bit code, split over 39 individual drives. Subsequent failure of any one of the 39 drives would not impair reading of the data, since the ECC code allows any single-bit error to be detected and corrected. [1] The (39,32) code is a SECDED-variant Hamming code, meaning it can detect and correct any single-bit error as well as detect any two-bit error.
Although operation is possible with a single failed drive, three spare drives were available to replace failed units until they are repaired. The ECC codes permit 100% recovery of the data on any one failed disk, allowing a new copy of this data to be reconstructed and written onto the replacement disk. Once this recovery is complete, the data base is considered to be healed. [1]
In today's terminology this would be labeled a RAID-2 subsystem. [2] However, these units shipped before the label RAID was formed.
The DataVault was an example of unusual industrial design. Instead of the usual rectilinear box, the cabinet had a gentle curve that made it look like an information desk or a bartender's station. [2]