Low Latency

Hiperspace was designed from the outset to offer the low latency possible because it provides head room for other changes that significantly reduce application complexity.

In tests, Hiperspace performs four times faster than a relational database with direct file-system access and twice as fast when remote, it achieves this by eliminating layers of transformation:

• protobuf encoding is used to avoid the bloat of key names in JSON and XML serialization
• The first object serialization, is the last serialization: objects are stored directly
• Keys are transformed from length prefix value to value prefix length in the client library to enable partial match of keys (e.g. from entity to segment)
• Indexes for alternate access paths (e.g. Child to Parent in a hierarchy) are created to avoid slow search
• Version timestamps are transformed so that latest versions are always found first, and older versions can be skipped in storage search
• SQL Query only where needed for reports/notebook
• ODATA / JSON / XML view only needed for Business Intelligence
• Where remote access is needed for parallel access a C++ gRPC is used