Today, we welcome the CEO of NeuroBlade, Elad Sity, as he discusses all things data analytics, and what organizations that have lots of data can do to overcome what is fast becoming a significant inhibitor to growth - the data analytics gap.

VMblog:  Elad, please tell the readers a bit more about yourself before we delve into the details of data analytics.

Elad Sity:  Thanks for having me. I'm the Founder and CEO of NeuroBlade, and our mission is to innovate at the infrastructure layer to create a new standard for how data analytics is done - making it faster and cheaper through more acceleration and better performance.

My co-founder and CTO, Eliad Hillel, and I are graduates of a military intelligence tech unit of the Israel Defense Forces. I was also a senior executive at SolarEdge, a global supplier of smart energy solutions, and one of the largest public companies in Israel today, before establishing NeuroBlade in 2018. My background is in building systems that combine hardware and software to generate a bigger value prop by leveraging both worlds.

So, technology is a huge part of who we are, and it has helped shape our vision for NeuroBlade.

VMblog:  So, what is NeuroBlade, and what is it you do?

Sity:  I mentioned the data analytics gap, and that's a good place to start.

Data is growing exponentially. I think that has been and is being discussed widely. Let's go back to why we need this data. We need it for insights, we need it for better businesses, better life and so on. But for that - we need to process all this data. Unfortunately, there's a gap in the computing resources required to process this data effectively and the amount of data collected and this gap just keeps getting bigger and bigger.

We all own a phone. When the first iPhone was released, it came with 16GB memory. The recent model ships with 1TB of memory. So, in just 14 years, the phone's memory increased 64 times, roughly doubling every two years. Now imagine a world full of people using phones that generate these massive amounts of data, and that's not even taking into account all the "meta" data that is being generated from location tracking, click stream data, etc. which is collected by other organizations through our phone usage.

Organizations have to extract insights from this data to make sense of it. The aim is to not only generate more revenue but also make smarter decisions to innovate and grow - which requires systems that are capable of analyzing data faster than ever before.

The latest research suggests that as much as 2/3 of the data which organizations have access to is not analyzed and in many situations, IT's solution is just throwing more servers at the problem, hoping it will help. Guess what? It doesn't.

And this is where NeuroBlade comes in.

We optimize the entire data analytics journey, end to end, thinking vertically from software and compute to the network and storage layers. We're thinking cloud native from the beginning, by adding an acceleration layer between the compute and storage which lets you get to a vertically integrated, purpose built approach for the analytics workload. Our approach is open to any database, any cloud, any storage. We call this hypercompute for analytics - the next generation of compute systems, for analytics.

VMblog:  Can you put this into context? Practically, how much can you really improve the analytics cycle?

Sity:  Using the approach we've developed, we see orders of magnitude of improvement in analytics performance and TCO, to the tune of 10-100x improvement in price performance over the fragmented approach that standard systems stick to. This is using our current technology generation, but we see even better numbers ahead. If I were looking at the improvements in traditional "piecemeal" approaches, it would be at best, 10%-30% incremental improvement, and sometimes in a real life scenario, not even that.

VMblog:  So, tell us more about this. What is the 'traditional' approach to data analysis, and why can't it keep up?

Sity:  The industry has been approaching the data analysis gap from different angles.

Database software providers are looking to optimize data analytics using smarter indexing or querying techniques or even just better code.

Cloud and infrastructure providers, including hyperscalers, have been optimizing either a CPU-based or GPU based approach to run that database software faster. For example, AWS Graviton processors based on ARM can now improve the price-performance curve for database companies like Databricks and Snowflake.

From the networking side, we see smart NICs and DPU sort of solutions, to improve the data flow in the system. And lastly, storage vendors are innovating faster storage solutions and smart storage solutions to try and speed up closer to the data.

For example, a smart SSD can do a specific kernel 1000x faster, but in real life scenarios, it may not even give a 10% boost to the workload, because of the knock-on effect on the rest of the data analytics supply chain.

The bottom line is: we're all trying to make each layer of the stack more computationally powerful and more intelligent. But to achieve the "best of all possible worlds" , optimizing each piece by 10%-20% will not get you there. You need to look at the entire stack end to end and make it all work together and most importantly seamlessly.

VMblog:  How is NeuroBlade addressing this problem?

Sity:  Our solution is an acceleration layer that sits between compute and storage, called the hardware-enhanced query system or HEQS. Our vision is that this works with any database, any cloud, any storage.

The HEQS snaps into the core analytical engine software, such as Spark, Trino, or any other DBMS system by using our API, and then it automatically begins to offload queries that HEQS can execute to run on our acceleration appliance instead of on your existing infrastructure.

The HEQS system is based on our HEQS processing unit, which is a processor we invented and is where a lot of the magic happens. It is tailored for analytical workloads with high I/O and high memory needs, and when integrated in our appliance solution, along with the networking and storage elements, it enables the orders of magnitude faster processing we discussed.

Now, there are many hardware elements in the system. We don't expect the user to know how to extract the performance from all of these. Our software stack orchestrates all the hardware pieces together, making sure there are no bottlenecks in the system and every component is utilized to the fullest. And, as promised, it is all seamless for the user.

VMblog:  What can analyzing data faster yield in the future?

Sity:  Let's just take a couple of examples.

During the pandemic, biotech companies used data for everything from tracking the spread of the virus to developing new drugs. To support data analytics like this, a powerful data analytics system, which reduced the analysis time by 35%, was built. Imagine what would have happened if we could speed that up even more.

Another example from a different domain, which we have a very big customer in, is semiconductor manufacturing. And by the way, this actually applies to any manufacturing facility. There is so much data generated and gathered at the manufacturing floor, and allow me to quote the customer here - "It takes me a full week to understand what happened today and what I have to fix". If we could shorten the 7 days to a few hours, this would allow for better yield, quality and of course quantity, all translate to a bigger and more profitable business.

Simply put, hours of work are reduced and executed in minutes. In 2023, that will be game changing for many fields and businesses.

VMblog:  And finally, what is your vision for hypercompute for analytics?

Sity:  We're at the beginning of an industry-wide recognition that we need to address the infrastructure that supports analytical workloads.

We can't keep throwing servers at the problem - per Einstein, that would just fall into the definition of insanity.

Imagine you could do everything in your day, 100 times faster - the possibilities are endless.

NeuroBlade does this for data analytics, speeding up data processing by 100 times and it opens up a new world of data exploration and it will be exciting to see the things companies and people will do with it.