The Faster Paradigm - Why Software is the Smarter Way
A wonderful human ability is to make split-second decisions and derive an action from them. The decision as such usually produces an experience. And depending on whether it was a "good" or "bad" decision, man is quite capable of looking back and deriving new options for action from these experiences. This is called "learning through action."
In IT today, we are confronted with diverse and challenging requirements.
A few years ago it was tough enough to deal with data availability and data security. Today IT managers have to cope with higher expectations for 100% availability, growing cyber security threats, more stringent requirements for high performance and low latency, as well as a shift from time-critical to real-time business requirements.
Today’s exponential data growth, generates new challenges to the workforce and infrastructure that would have been unthinkable a few years ago. We realized early on that you can't solve these problems by simply throwing more "faster and better" hardware at them. The solution was to find a smarter approach to help our customers implement a sustainable process for their businesses!
For that, we developed Neural Cache, a true deep learning software that works at the heart of all Infinidat systems. This enables our systems to take actual, software-based courses of action. This means that an InfiniBox actually learns over time and continuously adapts to specific data patterns and user behaviour. To enable this behavior, we use a trie structure, which, unlike classical structures such as a hash table, proves to be superior and extremely performant in multi-petabyte environments. Anyone who has entered a search term in the Google search engine and is surprised by the various search suggestions has experienced a trie structure in use.
We analyze the incoming data and their query patterns and ensure that the relevant data and potentially additional data are always available in the fastest memory area - the DRAM - so that the user can expect the fastest possible response times.
Here’s how it works:
If you imagine a bowl of grapes and you reach for some, you typically pull out other grapes hanging from the same vine. It is similar to Neural Cache. Related data is discovered, analyzed and loaded accordingly into our - up to 3TB large - DRAM memory, which is many times faster than the fastest flash media.
But real deep learning also means more. The results - measured here in cache hits or cache misses - are also analyzed again and, based on the results, an Infinidat system continues to optimize itself over time and actually learns. The clever thing about this solution is that it also works with changing data patterns. So if a customer's application and/or user behaviour changes, our Neural Cache learns something new and adapts it to the new circumstances.
The Neural Cache is placed in front of the physical memory as an abstraction layer. The beauty of this is that now the memory components remain independent of the software layer and you get an architecture that adapts to and incorporates each - or at best all - components. With the latest addition to our product portfolio - the InfiniBox SSA - we have demonstrated this impressively.
And this brings us back to the analogy from the beginning. Deep learning at Infinidat means permanent learning AND learning from what you have learned = learning by using different options for action and recognizing the best result.
That sounds all well and good, but what does that mean for our customers in real life? It means having a process that on the one hand gives them high flexibility, fast adaptability to changing requirements while also delivering higher speeds than even conventional flash systems can do.
Ultimately this means more throughput, faster response times and thus increased productivity. Customer satisfaction and internal acceptance are the logical consequence.