Controlling Data Egress to Reduce Cloud Computing Costs – Adit Madan, Alluxio

This is Techstrong tv. Hey guys. Thanks for the throw.

We are here with Audit Madan, who is director of product for alexio, and we're talking about one of the nastier little surprises in the world of the cloud. They're called Egress Fees Audit. Welcome the show.

Hi, Michael. So people have been complaining about egress fees for a while, but you know, they're kind of like taxes. Everybody just kind of sucks it up and deals with it.

But I guess the question I would have for you is, is there something to be done about this? How do we kind of minimize those costs? Uh, yeah, Michael, so egress fees, uh, as you know, is, is something that's incurred when there is cross region traffic or any traffic which is traversing outside the network of a cloud provider.

So these days, I think there's a lot of techniques that people are employing, uh, to, to avoid egress fees. Uh, one of the things that people typically do is they, they bear the operational cost of, of copying data manually. So, so that they don't have to access the data repeatedly across, across regions or across silos, which may exist in the cloud.

So I, I think making the process of sharing across regions or even cloud providers is, is something, uh, that, uh, needs to be tackled and not for everyone. Uh, I think it, it really depends on the scale of the organization, uh, when, who see these problems. But, uh, there, there are solutions out there which ma, which, uh, make this easy, uh, for people to manage, make sure that they're not copying data and accessing it repeatedly, uh, in including, uh, the, the solution that we've built here at El Alexia.

Are we encountering this issue more often now because there's so much more data flowing across those networks? Is that part of the conversation and why? Because it's been around as a topic for a while, but to your point, I think maybe more people are starting to feel the pain.

Yeah, I, I think, uh, Mike, one of the trends recently that's been there is, uh, with, uh, all of the hype around, uh, AI and machine learning and, and the shortage of GPS in the market. Uh, that is one, uh, one reason why, uh, this cross region traffic or even cross cloud traffic ha has become, uh, more of, uh, has become more substantive these days. Uh, so I think one of the trends that we are seeing is oftentimes, uh, folks have the right configuration of GPUs, let's say a available in, in, in a Microsoft cloud or, or available on-premises, but their data is spread across, uh, different locations.

And when you're training models which need, uh, repeated access to a lot of data, uh, that's one reason for, for data and the, and the, the co framework, which is computing on that data to be naturally separated. So I think that's why it's, it's becoming, uh, becoming a topic of interest, uh, to many more than it used to be. So what are the methods for controlling that?

I mean, and there's probably a few, but the one that you guys have, how does that actually work? Yeah, so, so the method that we have, uh, we've, uh, built, uh, a highly distributed, uh, data access layer, which sits between, uh, different computer engines and, and different storage types. So in, in the scenario that we were talking about, let's say you have a data or data in cloud A, what you want to process it in cloud B, uh, or you want to process it on premises, uh, we would, uh, deploy our software close to where the compute sits.

Uh, and, and we, one of the things we provide is, is a highly distributed, uh, uh, scalable cash. Uh, so, so caching is, is actually a, a very effective mechanism, uh, for, uh, different analytical and machine learning workloads in which, uh, we provide access to data, which is remote. Uh, so from, uh, cloud A to cloud B, uh, and cloud B is able to access it directly, uh, from cloud A without necessarily making a copy of that data from cloud A to cloud B and, and then making it available to, to competition in cloud B.

So, so it's in, in theory, in essence, it acts similar, uh, to, to the, to the workaround that I was mentioning earlier in which people manually copy data afro across the cloud, so that repeated access don't bear the cost, uh, the egress fees associated with that. The difference in the solution that we provide is, uh, the granularity, uh, uh, at which we move, uh, or the granularity at which we cash, uh, is, is fairly, uh, granular. So, so we minimize the amount of data which actually moves across the network.

Hmm. Cloud service providers are kind of fond of the fees that they generate. So are they okay with what you guys are doing?

That, that, that's a great question. So, uh, I, I think, uh, one of the things that I've learned by talking to all of the cloud providers is, uh, they, even though they may lose money in their short term, uh, but they, the number one thing that they care about is whatever is most suitable for their customers. So, because in the long run, if their customers are happy with the solutions or the services that they're getting from the cloud provider, they will make money over time.

So, uh, whatever is best for the customer, uh, is best for the cloud provider as well. E even though it may seem conflicting in, in the, in the near term. So the cloud service providers don't want to have the customers feel like they're being nickled and dimed, per se, when they're really making more money by consumption of compute instances.

Compute instances, exactly. So, so, uh, in, in these cases, uh, compute in instances, uh, tends to be one of the more, uh, most expensive, uh, line items in, in, in the bill, uh, that any, uh, enterprise might have with a cloud provider. So, yeah, like, like you said, uh, ni nickling, uh, nickel and dime, uh, that's not the business that most most cloud providers are in.

Hmm. Do you think that there's a greater sensitivity to cloud costs these days? Because the uncertain economic times we live in, I know one's quite sure whether we're in a recession or not, but we're all feeling a pinch somewhere.

Yeah, I, I think o overall, uh, the cost, uh, in the cloud, it has been, uh, a priority for enterprises in, in the past few years, uh, especially in, let's say the, the last five, seven years when a lot of large enterprises made the shift from their on-prem, on-prem data lakes and data warehouses into the cloud. I think right now what's happening is, as the volume, uh, of consumption of the cloud has really picked up, uh, some of the, the fees, uh, are, it's, it's aggregating over time, uh, and, and, uh, especially with the economic climate these days, uh, there's, there's more sensitivity around optimizing, uh, the cost in, in the cloud. So, uh, so percentage wise, if, if there's something which can improve the, your cloud consumption cost by like, let's say a 20 or, or 30%, just because the volumes of, of, uh, in the cloud are so significant now, uh, it's, it's, it's a fairly, uh, fairly significant, uh, saving, uh, for a company.

But, but again, like savings, um, in, in the cloud, uh, it always needs to be, uh, traded off with, um, just, uh, agility and, uh, all and operational, uh, ease, uh, as well. So, uh, at, even though cost is extremely important, at the same time, making sure that, uh, the enterprises are using the most out of, uh, getting the maximum agility, they're solving the most problem in the, in the cloud using the best services. Uh, there, there's a fine, uh, trade off, uh, but that, that's top of mind for, for a lot of organization that we speak to.

Do you think, is it really the reduction of the cost that people are after? I know that's probably, you know, somebody's goal, but it seems to me when I talk to people that are really more concerned about just trying to make it more predictable and not have these kind of spiky or irregular costs, and so they can plan better. So how much of it is cost reduction and how much of it is just simply better planning?

Uh, I, I, I think, uh, predictability has, has been the case, uh, for, for some time now. I, I think, uh, I think it's a mix actually. Like when, when I speak to customers, I do hear directly that even optimizing the cost in addition to being able to plan better is, is of importance to them.

And that's maybe more so the case recently, uh, with the economic climate, like we were talking about, but not so much, uh, before that, where just being able to predict better and plan better, uh, was, was top of mind recently I've seen a little bit of a shift in, in which cost optimization is, is also something, uh, which has, uh, ha has become, uh, a priority. Maybe not the number one priority, but definitely, definitely a priority. Do you think someday, uh, we might be applying some form of AI to reign in these costs?

Because the environments are somewhat complex and there's a lot of things that are changing? Definitely. I, I think we, we've already seen seen signs of, uh, AI being used for observability, which is kind of the first step of automating all of this.

So if you look at the different services providers that enterprise have available to them, they all have, have different cost characteristics as well. So, for example, storage might be cheap in one cloud, uh, but compute and, and GP access might be cheaper in another cloud. So how do ma manage the balance dynamically for an organization?

Um, and especially with, with the volumes of data that we have picking up right now, I, I, I, I think AI definitely has a place in, in this, uh, it's, it's, it's still, there's still a little way, uh, to go to really, uh, for this market, uh, in, in terms of the, the tool set, uh, which is solving the, to, to mature to an extent at which in which it can become fully automated. Uh, but I, I think we are definitely headed in that direction. Moving work loans from one cloud to another is a major undertaking.

But do you think organizations are maybe gonna be a little more particular about what cloud they use based on the total cost? Cause seems like historically it was just kinda up to the developer's preference. Yeah, I, I, I think what's happening these days is the split that we see across our customers is it's really based on which cloud is most suitable for the service at hand.

Uh, so if, uh, certain services are more suitable for, for cer a certain cloud and certain services are more suitable for another, so, uh, the, the trend that we've seen is if you just look across the, the data pipeline, uh, that some people are managing, there could, there are portions of a pipeline, uh, which are residing in one cloud, whereas there's a second portion of it, which is consuming the output of, uh, processing which had, which happened in, in, in the first cloud. So I think it, it's, like I said, it's still maturing, uh, and it's not easy enough just yet, uh, for it to be mainstreamed. So we've only seen, um, the, the, the highly skilled, uh, big tech leaders and extremely, uh, savvy, uh, enterprise customers go down that route so far.

Uh, but, uh, as this market matures, as it becomes easier to use, uh, I think it, it's only going to pick up. So what's your best advice to folks? I think they struggle with trying to get their arms around something that has been either not managed or quasi managed for a while.

So how do you get started? Yeah, I, I think, uh, the, the biggest, uh, so there's always a trade off right now between how, uh, out of box, uh, do you get things, uh, and how much control, uh, do you have over it? So the, the balance that we've seen really depends on how equipped, uh, teams are.

So, uh, what I mean by that is, uh, the more that you're, you, you're able to do things on your own, uh, which in certain cases might mean that it'll take a little longer to get started. Uh, the more, the more customizable the system is, and the more customizable the system is, the lower the cost is going to be because you're going to be able to customize, uh, the stack that you're using to your specific needs. So, uh, what the, the strategy that I would, uh, I would employ, uh, as an enterprise would be to, to be able to migrate a across the solutions which are most suitable for you a at a given point in time, and, and realize the fact that, uh, uh, the service which is most appropriate for you might change my, might change over time.

Uh, so for, uh, for example, if your team is not, uh, really needs to get started on some key initiatives, and the, the most important thing right now is not cost at all. The most important thing is to get the initiative, uh, uh, going as soon as possible, use the most easiest cloud, uh, service out there, uh, but in, in the background as, uh, as that matures, uh, slowly start migrating your solutions to things that you are familiar with, to a stack that you can operate, which is more customizable and something that you can control the cost over time. So at the same time there, there's a balance between getting going and then maintaining the cost cost over time.

And I think with that mentality in mind, uh, that the right solution for you might shift over time. Uh, I, I think it, it can, uh, solve a lot of problems for enterprises in the long term, especially with respect to, to managing costs as well. All right, folks, you're hearing it here.

Hey, it may feel good when you initially get started. Just remember there's always a bill to come at the end of the party audit. Thanks for being on the show show.

Thank you so much, Michael. All right. Back to you guys in the studio.