Each storage node is virtualized and its capacity is allocated to a volume.
A gridstore can contain multiple volumes and storage nodes can provide resources to multiple volumes.
A Gridstore volume appears as a standard Microsoft Windows Network Share with a drive letter to end users
and applications.
Write Files
When an application writes a file or a user copies a file to a Gridstore volume,
Gridstore does three things.
1) Redundancy is calculated. The number of redundant storage nodes is a configurable number of nodes
from the total number of storage nodes in a volume.
2) The file is then sliced into fragments equal to the number of storage nodes in the volume.
3) The slices are written to the grid in parallel.
Read Files
When an application reads a file, the following process takes place:
1) Gridstore sends a request in parallel to the storage nodes in the volume.
2) Each node responds in parallel sending any file slices in holds.
3) The Gridstore client calculates any missing slices and recombines the slices back into the original file.
Benefit of this Approach - No Wasted Resources
Redundancy not Replication - Gridstore uses redundancy calculations similar to RAID.
Replication copies the entire file sometimes multiple times - this can generate overheads of 200-300%
depending on the number of replicas. Redundancy adds overhead of 20-30% and provides a higher level
of reliability.
Parallel Processing - The high levels of reliability with Gridstore
are only possible due to the amount of
processing power that can be aggregated from the client machines. Simply put, Gridstore can aggregate the
same amount of processing power from 25 PCs as a multi-million dollar SAN. By utilizing what you have
already paid for, Gridstore is able to offer levels of reliability not possible with enterprise storage systems.
Parallel I/O - After the expensive processing is completed at the client,
data is sent in parallel to
each storage node. The more storage nodes, the less work each one does and the greater the data path
to your storage. Instead of creating network bottlenecks at the server, Gridstore uniformly spreads the
load across the network to maximize network efficiency.
