We'll learn how bringing analytics to a cloud services model also allows smaller and less data-architecture-experienced firms to benefit from the latest in big-data capabilities. And we'll explore how Dasher Technologies is helping to usher in this democratization of big data value to more players in less time.
Listen to the podcast. Find it on iTunes. Get the mobile app. Read a full transcript or download a copy.
To share how a fast ramp-up for big data as a service has evolved, we're joined by Justin Harrigan, Data Architecture Strategist at Dasher Technologies, as well as Chris Saso, Senior Vice President of Technology at Dasher Technologies in Campbell, California. The discussion is moderated by me, Dana Gardner, Principal Analyst at Interarbor Solutions.
Here are some excerpts:
Gardner: Justin, how have big-data practices changed over the past five years to set the stage for rapid leveraging of big-data capabilities?
Harrigan: Back in 2010, we saw big data become mainstream. Hadoop became a household name in the IT industry, doing scale-out architectures. Linux databases were becoming common practice. Moving away from traditional legacy, smaller, slower databases allowed this whole new world of analytics to open up to previously untapped resources within companies. So data that people had just been sitting on could now be used for actionable insights.
Today, we have approachable analytics, analytics as a service, hardened architectures that are almost turnkey with back-end hardware, database support, and applications -- all integrating seamlessly. As a result, the user on the front end, who is actually interacting with the data and making insights, is able to do so with very little overhead, very little upkeep, and is able to turn that data into business-impact data, where they can make decisions for the company.
Gardner: Justin, how big of an impact has this had? How many more types of companies or verticals have been enabled to start exploring advanced, cutting-edge, big-data capabilities? Is this a 20 percent increase? Perhaps almost any organization that wants to can start doing this.
Harrigan: The tipping point is when you outgrow your current solutions for data analytics. Data analytics is nothing new. We've been doing it for more than 50 years with databases. It’s just a matter of how big you can get, how much data you can put in one spot, and then run some sort of query against it and get a timely report that doesn’t take a week to come back or that doesn't time out on a traditional database.
For people dipping their toes in the water for their first larger dataset analytics, there's a whole host of avenues available to them. They can go to some online providers, scale up a database in a couple of minutes, and be running.
They can download free trials. HP Vertica has a community edition, for example, and they can load it on a single server, up to terabytes, and start running there. And it’s significantly faster than traditional SQL.
It’s much more approachable. There are many different flavors and formats to start with, and people are realizing that. I wouldn’t even use the term big data anymore; big data is almost the norm.
Gardner: I suppose maybe the better term is any data, anytime.
Harrigan: Any data, anytime, anywhere, for anybody.
Gardner: I suppose another change over the past several years has been an emphasis away from batch processing, where you might do things at an infrequent or occasional basis, to this concept that’s more applicable to a cloud or an as-a-service model, where it’s streaming, continuous, and then you start reducing the latency down to getting close to real time.
Are we starting to see more and more companies being able to compress their feedback, and start to use data more rapidly as a result of this shift over the past five years or so?
Harrigan: It’s important to address the term big data. It’s almost like an umbrella, almost like the way people use cloud. With big data, you think large datasets, but you mentioned speed and agility. The ability to have real-time analytics is something that's becoming more prevalent and the ability to not just run a batch process for 18 hours on petabytes of data, but having a chart or a graph or some sort of report in real time. Interacting with it and making decisions on the spot is becoming mainstream.
We did a blog post on this not long ago, talking about how instead of big data, we should talk about the data pipe. That’s data ingest or fast data, typically OLTP data, that needs to run in memory or on hardware that's extremely fast to create a data stream that can ingest all the different points, sensors, or machine data that’s coming in.
Then we've talked about smarter analytic data that required some sort of number-crunching dataset on data that was relevant, not data that was real-time, but still fairly new, call it seven days or older and up to a year. And then, there's the data lake, which essentially is your data repository for historical data crunching.
Those are three areas you need to address when you talk about big data. The ability to consume that data as a service is now being made available by a whole host of companies in very different niches.
It doesn’t matter if it’s log data or sensor data, there's probably a service you can enable to start having data come in, ingest it, and make real-time decisions without having to stand up your own infrastructure.
Gardner: Of course, when organizations try to do more of these advanced things that can be so beneficial to their business, they have to take into consideration the technology, their skills, their culture -- people, process and technology, right?
Chris, tell us a bit about Dasher Technologies and how you're helping organizations do more with big-data capabilities, how you address this holistically, and this whole approach of people, process and technology.
Saso: Dasher was founded in 1999 by Laurie Dasher. To give you an idea of who we are, we're a little over 65 employees now, and the size of our business is somewhere around $100 million.
We started by specializing in solving major data-center infrastructure challenges that folks had by actually applying the people, process and technology mantra. We started in the data center, addressing people’s scale out, server, storage, and networking types of problems. Over the past five or six years, we've been spending our energy, strategy, and time on the big areas around mobility, security, and of course, big data.
As a matter of fact, Justin and I were recently working on a project with a client around combining both mobility information and big data. It’s a retail client. They want to be able to send information to a customer that might be walking through a store, maybe send a coupon or things like that. So, as Justin was just talking about, you need fast information and making actionable things happen with that data quickly. You're combining something around mobility with big data.
Dasher has built up our team to be able to have a set of solutions that can help people solve these kinds of problems.
Gardner: Justin, let’s flesh that out a little bit around mobility. When people are using a mobile device, they're creating data that, through apps, can be shared back to a carrier, as well as application hosts and the application writers. So we have streams of data now about user experience and activities.
We also can deliver data and insights out to people in the other direction in that real-time of fashion, a closed loop, regardless of where they are. They don’t have to be at their desk, they don’t have to be looking at a specific business-intelligence (BI) application for example. So how has mobility changed the game in the past five years?
Harrigan: Dana, it’s funny you brought up the two different ways to capture data. Devices can be both used as a sensor point or as a way to interact with data. I remember seeing a podcast you did with HP Vertica and GUESS regarding how they interacted with their database on iPads.
In regards to interacting with data, it has become not only useful to data analysts or data scientists, but we can push that down into a format so lower-level folks who aren't so technical. With a fancy application in front of them, they can use the data as well to make decisions for companies and actually benefit the company.
You give that data to someone in a store, at GUESS for example, who can benefit by understanding where in the store to put jeans to impact sales. That’s huge. Rather than giving them a quarterly report and stuff that's outdated for the season, they can do it that same day and see what other sites are doing.
On the flip side, mobile devices are now sensors. A mobile device is constantly pinging access points over wi-fi. We can capture that data and, through a MAC address as an unique identifier, follow someone as they move through a store or throughout a city. Then, when they return, that person’s data is captured into a database and it becomes historical. They can track them through their device.
It allows a whole new world of opportunities in terms of the way retailers interact with where they place merchandise, the way they interact with how they staff stores to make sure they have the proper amount of people for the certain time, what weather impact has on the store.
Lastly, as Chris mentioned, how do we interact with people on devices by pushing them data that's relevant as they move throughout their day?
The next generation of big data is not just capturing data and using it in reports, but taking that data in real time and possibly pushing it back out to the person who needs it most. In the retail scenario, that's the end users, possibly giving them a coupon as they're standing in front of something on a shelf that is relevant and something they will use.
Gardner: So we're not just talking about democratization of analytics in terms of the types of organizations, but now we're even talking about the types of individuals within those organizations.
Do you have any examples of some Dasher’s clients that have been able to exploit these advances and occurrences with mobile and cloud working in tandem, and how that's produced some sort of a business benefit?
Harrigan: A good example of a client who leveraged a large dataset is One Kings Lane. They were having difficulty updating the website their users were interacting with because it’s a flash shopping website, where the information changes daily, and you have to be able to update it very quickly. Traditional technologies were causing a business impact and slowing things down.
They were able to leverage a really fast columnar database to make these changes and actually grow the inventory, grow the site, and have updates happen in almost real time, so that there was no impact or downtime when they needed to make these changes. That's a real-world example of when big data had the direct impact on the business line.
Gardner: Chris, tell us a little bit about how Dasher works with Hewlett Packard Enterprise technologies, and perhaps even some other HP partners like GoodData, when it comes to providing analytics as a service?
Saso: HP has been a longtime partner from the very beginning, actually when we started the company. We were a partner of Vertica before HP purchased them back in 2011.
We started working with Vertica around big data, and Justin was one of our leads in that area at the time. We've grown that business and in other business units within HP to combine solutions, Vertica, big data, and hardware, as Justin was just talking about. You brought up the applications that are analyzing this big data. So we're partners in the ecosystem that help people analyze the data.
Once HP Vertica, or what have you, has done the analysis, you have to report on that and make it in a nice human-readable form or human-consumable form. We’ve built out our ecosystem at Dasher to have not only the analytics piece, but also the reporting piece.
Gardner: And on the as a service side, do you work with GoodData at all or are you familiar with them?
Saso: Justin, maybe you can talk a little bit about that. You've worked with them more I think on their projects.
Optimizing the environment
Harrigan: GoodData is a large consumer of Vertica and they actually leverage it for their back-end analytics platform for the service that they offer. Dasher has been working with GoodData over the past year to optimize the environment that they run on.
Vertica has different deployment scenarios, and you can actually deploy it in a virtual-machine (VM) environment or on bare-metal. And we did an analysis to see if there was a return on investment (ROI) on moving from a virtualized environment running on OpenStack to a bare-metal environment. Through a six-month proof of concept (POC), we leveraged HP Labs in Houston. We had a four-node system setup with multiple terabytes of data.
We saw 4:1 increase in performance in moving from a VM with the same resources to a bare-metal machine. That’s going to have a significant impact on the way they move data in their environment in the future and how they adjust to customers with larger datasets.
Gardner: When we think about optimizing the architecture and environment for big data, are there any other surprises or perhaps counter-intuitive things that have come up, maybe even converged infrastructure for smaller organizations that want to get in fast and don’t want to be too concerned with the architecture underlying the analytics applications?
Harrigan: There's a tendency now with so many free solutions out there to pick a free solution, something that gets the job done now, something that grows the business rapidly, but to forget about what businesses will need three years down the road, if it's going to grow, if it’s going to survive.
There are a lot of startups out there that are able to build a big data infrastructure, scale it to 5,000 nodes, and then they reach a limit. There are network limits on how fast the switch can move data between nodes, constantly pushing the limits of 10 Gbyte, 40 Gyte and soon 100 Gbyte networks to keep those infrastructures up.
Depending on what architecture you choose, you may be limited in the number of nodes you can go to. So there are solutions out there that can process a million transactions per second with 100 nodes, and then there are solutions that can process a million transactions per second with 20 nodes, but may cost slightly more.
If you think long-term, if you start in the cloud, you want to be able to move out of the cloud. If you start with an open ecosystem, you want to make sure that your hardware refresh is not going to cost so much that the company can’t afford it three years down the road. One of the areas we help consult with, when picking different architectures, is thinking long-term. Don't think six weeks down the road, how are we going to get our service up and running? Think, okay, we have a significant client install base, how we are going to grow the business from three to five years and five to 10 years?
Gardner: Given that you have quite a few different types of clients, and the idea of optimizing architecture for the long-term seems to be important, I know with smaller companies there’s that temptation to just run with whatever you get going quickly.
What other lessons can we learn from that long-term view when it comes to skills, security, something more than the speeds and feeds aspects of thinking long term about big data?
Harrigan: Think about where your data is going to reside and the requirements and regulations that you may run into. There are a million different regulations we have to do now with HIPAA, ITAR, and money transaction processes in a company. So if you ever perceive that need, make sure you're in an ecosystem that supports it. The temptation for smaller companies is just to go cloud, but who owns that data if you go under, or who owns that data when you get audited?
Another problem is encryption. If you're going to start gaining larger customers once you have a proven technology or a proven service, they're going to want to make sure that you're compliant for all their regulations, not just your regulations that your company is enforcing.
There's logging that they're required to have, and there is going to be encryption and protocols and the ability to do audits on anyone who is accessing the data.
Gardner: On this topic of optimizing, when you do it right, when you think about the long term, how do you know you have that right? Are there some metrics of success? Are there some key performance indicators (KPIs) or ROIs that one should look to so they know that they're not erring on the side of going too commercial or too open source or thinking short term only? Maybe some examples of what one should be looking for and how to measure that.
Harrigan: That’s going to be largely subjective to each business. Obviously if you're just going to use a rule of thumb, it shouldn't cost you more money than it makes you. If you implement a system and it costs you $10 million to run and your ROI is $5 million, you've made a bad decision.
The two factors are the value to the business. If you're a large enterprise and you implement big data, and it gives you the ability to make decisions and quantify those decisions, then you can put a number to that and see how much value that big-data system is creating. For example, a new marketing campaign or something you're doing with your remote sites or your retail branches and it’s quantifiable and it’s having an impact on the business.
The other way to judge it is impact on business. So, for ad serving companies, the way they make money is ad impressions, and the more ad impressions they can view, for the least cost in their environment, the higher return they're going to make. The delta is between the infrastructure costs and the top line that they get to report to all their investors.
If they can do 56 billion ad impressions in a day, and you can double that by switching architectures, that’s probably a good investment. But if you can only improve it by 10 percent by switching architectures, it’s probably too much work for what it’s worth.
Gardner: One last area on this optimization idea. We've seen, of course, organizations subjectively make decisions about whether to do this on-premises, maybe either virtualized or on bare metal. They will do their cost-benefit analysis. Others are looking at cloud and as a service model.
Over time, we expect to have a hybrid capability, and as you mentioned, if you think ahead that if you start in the cloud and move private, or if you start private you want to be able to move to the cloud, we're seeing the likelihood of more of that being able to move back and forth.
Thinking about that, do you expect that companies will be able to do that? Where does that make the most sense when it comes to data? Is there a type of analysis that you might want to do in a cloud environment primarily, but other types of things you might do private? How do we start to think about breaking out where on the spectrum of hybrid cloud set of options one should be considering for different types of big-data activity?
Harrigan: In the large data analytics world, it’s almost an either-or decision at this time. I don’t know what it will look like in the future.
Workloads that lend themselves extremely well to the cloud are inconsistent, maybe seasonal, where 90 percent of your business happens in December. Seasonal workloads like that lend themselves extremely well to the cloud.
Or, if your business is just starting out, and you don't know if you're going to need a full 400-node cluster to run whatever platform or analytics platform you choose, and the hardware sits idle for 50 percent of the time, or you don’t get full utilization. Those companies need a cloud architecture, because they can scale up and scale down based on needs.
Companies that benefit from on-premise are ones that can see significant savings by not using cloud and paying someone else to run their environment. Those companies typically pin the CPU usage meter at 100 percent, as much as they can, and then add nodes to add more capacity.
The best advice I could give is, if you start in the cloud or you start on bare metal, make sure you have agility and you're able to move workloads around. If you choose one sort of architecture that only works in the cloud and you are scaling up and you have to do a rip and replace scenario just to get out of the cloud and move to on-premise, that’s going to be significant business impact.
One of the reasons I like HP Vertica is that it has a cloud instance that can run on a public cloud. That same instance, that same architecture runs just as well on bare metal, only faster.
Gardner: Chris, last word to you. For those organizations out there struggling with big data, trying to figure out the best path, trying to think long term, and from an architectural and strategic point of view, what should they consider when coming to an organization like Dasher? Where is your sweet spot in terms of working with these organizations? How should they best consider how to take advantage of what you have to offer?
Saso: Every organization is different, and this is one area where that's true. When people are just looking for servers, they're pretty much all the same. But when you're actually trying to figure out your strategy for how you are going to use big-data analytics, every company, big or small, probably does have a slightly different thing they are trying to solve.
That's where we would sit down with that client and really listen and understand, are they trying to solve a speed issue with their data, are they trying to solve massive amounts of data and trying to find the needle in a haystack, the golden egg, golden nugget in there? Each of those approaches certainly has a different answer to it.
So coming with your business problem and also what you would like to see as a result -- we would like to see x-number of increase in our customer satisfaction number or x-number of increase in revenue or something like that -- helps us define the metric that we can then help design toward.
Listen to the podcast. Find it on iTunes. Get the mobile app. Read a full transcript or download a copy. Sponsor: Hewlett Packard Enterprise.
You may also be interested in:
- How HTC centralizes storage management to gain visibility and IT disaster avoidance
- Big data, risk, and predictive analysis drive use of cloud-based ITSM, says panel
- Rolta AdvizeX experts on hastening big data analytics in healthcare and retail
- The future of business intelligence as a service with GoodData and HP Vertica
- Enterprises opting for converged infrastructure as stepping stone to hybrid cloud
- Redcentric orchestrates networks-intensive merger using advanced configuration management database
- HP pursues big data opportunity with updated products, services, developer program
- How eCommerce sites harvest big data across multiple clouds
- How Localytics uses big data to improve mobile app development and marketing
- HP hyper-converged appliance delivers speedy VDI and apps deployment and a direct onramp to hybrid cloud
- Full 360 takes big data analysis cloud services to new business heights