free counter
Tech

Podcast: Storage the main element bottleneck for AI processing

We speak to Panasas concerning the dependence on storage that may deal with delivering high volumes of small files for artificial intelligence with the throughput and latency had a need to service costly GPUs

Antony Adshead

By

Published: 01 Sep 2022

We speak to Jeff Whitaker, vice-president for product marketing at scale-out NAS maker Panasas, about why storage may be the key bottleneck in artificial intelligence (AI) processing.

In this podcast, we go through the key requirements of AI processing and how watching the storage it needs may bring benefits.

Whitaker discusses the necessity to get plenty of data into AI compute resources quickly and how some are tempted to throw compute at the issue. Instead, he argues that attention ought to be paid to storage which has the proper throughput and latency performance profiles to cope with plenty of small files, as within AI

Antony Adshead: Do you know the challenges organisations face with regards to storage for high-performance applications?

Jeff Whitaker: With regards to high-performance applications…the application form is trying to access results fast. Its looking to get to a choice, looking to get information back for the surroundings that is utilizing the application.

Theres ordinarily a heavy reliance on the compute side of the, and sometimes an over-reliance. Frequently which can be determined, which can be resolved by [asking], what [does] an average application environment appear to be? Its compute, its network and its own storage.

And I say storage third because often storage may be the very last thing thats considered on looking to get performance out of a credit card applicatoin environment.

Among the things we prefer to look at is, with regards to an application, do you know the data needs? What type of throughput is necessary, what type of latencies are needed, the facts likely to take for that application to perform as efficiently as you possibly can?

And frequently, customers and partners have viewed solving the task by throwing more compute at making the applications faster, but actually the bottleneck comes around storage.

Its very important to visitors to understand with regards to their environment they ought to consider the data needs before each goes and make an effort to solve the issue with just compute.

So, its a really matter of attempting to build a competent environment to have the results they want. They have to look at which kind of a storage environment can solve the challenges of these application.

Adshead: Do you know the key trends you’re seeing, particularly round the convergence of high-performance computing (HPC) with high-end enterprise storage, artificial intelligence and machine learning?

Whitaker: HPC has traditionally been a credit card applicatoin environment that requires lots of data. And lots of times, the storage environment must be something special that may scale and address the throughput so the compute doesnt just sit there idle. It requires lots of data to arrive there.

What weve began to see with the AI world and getting beyond just the development and discovering ideas, theyre essentially applications. An AI environment is wanting to process plenty of data and move on to an outcome, especially through the training process theres tonnes of data being pumped into compute. So, in this instance its often GPUs [graphics processing units] which are used and the ones are expensive no one really wants to sit there and also have those idle.

So, how fast it is possible to pump the info into an AI environment is crucial to how fast the application form can run or the AI training can run. In the event that you consider it, its almost on a par using what an HPC environment typically appears like where youre ingesting a tonne of data looking to get a result, which means you should look at what those data needs are for that training process or for the various kinds of HPC workloads and make an effort to solve the task from there.

The main one difference that people see here’s often within an HPC world, we see large files being pumped in to the compute. Whereas in the AI side, we see tonnes of smaller files being pumped in to the compute.

And actually the bottleneck becomes how fast is it possible to get that data in to the compute to get to an outcome.

And really venturing out there and saying can a normal enterprise storage environment solve that require for you personally?

Its latency, its throughput. Traditional environments be capable of have small latency, but looking to get very scalable throughput is quite challenging and thats whenever we begin to look at different kind of architecture like parallel solutions that may scale consistently, based on just how much performance you will need, really solving that challenge of ingesting tonnes of data into those compute environments.

Read more on SAN, NAS, solid state, RAID

Read More

Related Articles

Leave a Reply

Your email address will not be published.

Back to top button

Adblock Detected

Please consider supporting us by disabling your ad blocker