Yossi Weihs

turned the whole software stack that feeds information into the database into almost like a streaming process.

So as we collect information from the clusters, we are fine-tuning the ZAPI calls to batch

to batch up the information and get it in a clean format.

The performance stats, we grab those from files off the storage system, so there's really no zappy calls happening there.

And those get processed and written to the database.

But as we do that, we also run all of our analytics in parallel.

So we're not going through a process which was actually very prevalent two and three years ago in the product where we would grab the data, stick it in the database, then two minutes later, pull it out for analytics, throw it away.

Ten minutes later, pull it out for summarization, throw it away.

All of that has been optimized, and what customers at the end of the day will feel is that they need to put less resources against the product to get the same end result.

So this has been an ongoing pain for customers.

Let's face it this way.

It's been a historical pain.

There are a lot of customers that are still very concerned about this.

And we've actually gotten to the point where we have removed the limitation where we would not allow more than a single performance monitoring server to hit the cluster.

So back with ONTAP 8.3, we introduced a feature called the Performance Archiver in ONTAP, which allows ONTAP to basically automatically collect all the performance counters and put them in a nice, compact little file.

And all we do is basically read that file off the cluster every few minutes to get all the performance counters we need.

So, it really has actually collecting performance metrics has become more efficient than collecting some of the configuration settings.

Although, also with those, we benchmark those with ONTAP, the ONTAP team, and the ones that were taking a lot of time, we've sort of batched them and optimized them so that we run really well over high latency web links.

We're basically at a place where a highly centralized unified manager can happily monitor a cluster on the other side of the planet where a year or two ago, that wasn't really best of breed practice.

So that's a great question.