|Redis Adds TimeSeries And AI Support|
|Written by Kay Ewbank|
|Thursday, 04 April 2019|
Redis Labs has added two new data models to the models supported by the key-value store. The company has also added a novel programmability paradigm for multi-model operation.
Redis is an open source, BSD licensed, advanced key-value store where the keys can contain strings, hashes, lists, sets and sorted sets. It’s increasing in popularity for web development as a session state store because of its simplicity and rich data structure support.
The first new data model, RedisTimeSeries, is designed to collect and store high volume and velocity data, and organize it by time intervals. The aim is to make it easier to find useful data points, and to do this the time series model has built-in tools for downsampling, aggregation, and compression.
The second new data model is RedisAI. The developers have designed this to let analysts run AI data models within Redis. RedisAI is a Redis module for serving tensors and executing deep learning models. RedisAI makes use of a new tensor data type, along with new commands that let you get and set tensors from your deep learning client. Two more data types, Models and Scripts, have been added for model runtime features. The current preview has support for deep learning frameworks including TensorFlow, PyTorch, and TorchScript. The model uses Redis Cluster features on GPU-based servers, and the aim is to improve the speed with which analytics can be conducted and actions taken by reducing processing overheads.
The other announcement regarding the Redis data store is RedisGears. This is an in-database serverless engine based on the Redis Cluster distributed architecture. The developers say it's a dynamic execution framework for Redis that supports full Python syntax and has a low level C API. RedisGears supports both event-driven asynchronous access and transaction-based synchronous operations. The aim of RedisGears is to allow the user to build an operations pipe (OPP) that each key in Redis will pass through. Results from the first operation will pass as input to the second operation, results from the second operation will pass as input to the third operation, and so on. Results from the last operation will pass to the user as a reply. The pipe builds using a Python script and then runs in a background thread.
or email your comment to: email@example.com
|Last Updated ( Friday, 05 April 2019 )|