Listen to the podcast. Find it on iTunes. Get the mobile app for iOS or Android. Read a full transcript or download a copy.
To learn more about how big data is best architected for the largest web applications, BriefingsDirect sat down with Jimmy Mohsin, Principal Software Architect at Norjimm LLC, a consultancy based in Princeton, New Jersey. The discussion is moderated by me, Dana Gardner, Principal Analyst at Interarbor Solutions.
Here are some excerpts:
Gardner: How are large web applications deciding on the right big data architecture?
Mohsin: There's a lot of interest in trying to deal with large data volumes, not only large data volumes, but also data that changes rapidly. Now, there are many companies that have very large datasets, some in terabytes, some in petabytes and then they're getting live feeds.
The data is there and it’s changing rapidly. The traditional databases sometimes can’t handle that problem, especially if you're using that database as a warehouse and you're reporting against it.
Basically, we have kind of a moving-target situation. With HP Vertica, what we've seen is the ability to solve that problem in at least some of the cases that I've come across, and I can talk about specific use cases in that regard.
Gardner: Before we get into a specific use case, I'm interested particularly in some of these input/output issues. People are trying to decide how to move the data around. They're toying with cloud. They're trying to bring data for more types of traditional repositories. And, as you say, they're facing new types of data problems with streaming and real-time feeds.
How do you see them beginning this process when they have to handle so many variables? Is it something that’s an IT architecture, or enterprise architecture, or data architecture? Who's responsible for this, given that it’s now a rather holistic problem?
Mohsin: In my present project, we ran into that. The problem is that many companies don't even have a well defined data-architecture team. Some of them do. You'll find a lot of companies with an enterprise-architect role and you'll have some companies with a haphazard definition of an architectural group.
I know of a recent situation where we had to build a team of four people, and only one was an architect. But we built a virtual team of four people who were able to assemble and collate all the repositories that spanned 15 years and four different technology flavors, and then come up with an approach that resulted in a single repository in HP Vertica.
So there are no easy answers yet, because organizations just aren't uniformly structured.
Gardner: Well, I imagine they'll be adapting, just like we all are, to the new realities. In the meantime, tell me about a specific use case that demonstrates the intensity of scale and velocity, and how at least one architecture has been deployed to manage that?
Mohsin: One of my present projects deals with one of the world's largest retailers. It's eCommerce, online selling. One of the things they do, in addition to their transactions of buying and selling, is email campaign management. That means staying in touch with the customer on the basis of their purchases, their interests, and their profiles.
One of the things we do is see what a certain customer’s buying preferences have been over the past 90 days. Knowing that and the customer’s profile, we can try to predict what their buying patterns will be. So we send them a very tailored message in that regard. In this project, we're dealing with about 150 to 160 million emails a day. So this is definitely big data.
Here we have online information coming into one warehouse as to what's happening in the world of buying and selling. Then, behind the scenes, while that information is being sent to the warehouse, we're trying to do these email campaigns.
This is where the problem becomes fairly complicated. We tried traditional relational database management systems (RDBMS), and they kind of worked, but we ran into a slew of speed and performance issues. That's really where the big-data world was really beneficial. We were able to address that problem in about a seven-month project that we ran.
Gardner: And this was using HP Vertica?
Mohsin: We did an evaluation. We looked at a few databases, and the corporate choice was Vertica. We saw that there is a whole bunch of big-data vendors. The issue is that many of the vendors don't have any large organizations behind them, and Vertica does. The company management felt that this was a new big database, but HP was behind it, and the fact that they also use HP hardware helped a lot.
They chose Vertica. The team I was managing did a proof of concept (POC) and we were able to demonstrate that Vertica would be able to handle the reporting that is tied to the email campaign management. We ran a 90 day POC, and the results were so positive that there was an interest in going live. We went live in about another 90 days, following a 90-day POC.
Gardner: I understand that Vertica is quite versatile. I've heard of a number of ways in which it's used technically. But this email campaign problem almost sounds like a transactional issue, a complex event processing issue, or a transfer agent scaling issue. How does big data, Vertica, and analytics come to bear on this particular problem?
Mohsin: It's exactly what you say it is. As we are reporting and pushing out the campaigns, new information is coming in every half hour, sometimes even more frequently. There's a live feed that's updating the warehouse. While the warehouse is being updated, we want to report against it in real time and keep our campaigns going.
The key point is that we can't really stop any of these processes. The customers who are managing the campaigns want to see information very frequently. We can’t even predict when they would want their information. At the same time, the transactional systems are sending us live feeds.
The problem we ran into with the traditional RDBMS is that the reporting didn't function when the live feeds were underway. We couldn't run our back-end email campaign reports when new data was coming in.
One of the benefits Vertica has, due to its basic architecture and its columnar design is that it's better positioned to do that. This is what we were able to demonstrate in the live POC, and nobody was going to take our word for it.
The end user said, "Take few of our largest clients. Take some of our clients that have a lot of transactions. Prove that the reports will work for those clients." That's what we did in 30 days. Then, we extended it, and then in 90 days, we demonstrated the whole thing end to end. Following that was the go-live.
Gardner: You had to solve that problem of the live feeds, the rapidity of information. Rather going to a stop, batch process, analyze, repeat, you've gained a solution to your problem.
But at the same time, it seems like you're getting data into an environment where you can analyze it and perhaps extract other forms of analysis, in addition to solving your email, eCommerce trajectory issues. It seems to me that you're now going to have the opportunity to add a new dimension of analysis to what's going on and perhaps we find these transactions more toward a customer inference benefit.
More than a database
Mohsin: One of the things internally that I like to say is that Vertica isn't just a big database, it’s more than just a database. It's really a platform, because you have distributed all, you are publishing other tools. When we adopted it and went live with this technology, we first solved the feeds and speeds problem, but now we're very much positioned to use some of the capabilities that exist in Vertica.
We had Distributed R being one of them, Inference Analysis being another one, so that we can build intelligent reports. To date, we've been building those outside the RDBMS. RDBMS has no role in that. With Vertica, I call it more of a data platform. So we definitely will go there, but that would be our second phase.
As the system starts to function and deliver on the key use cases, the next stage would be to build more sophisticated reports. We definitely have the requirements and now we have the ability to deliver.
Gardner: Perhaps you could add visualization capabilities to that. You could make a data pool available to more of the constituents within this organization so that they could innovate and do experiments. That’s a very powerful stuff indeed.
Is there anything else you can tell us for other organizations that might be facing similar issues around real-time feeds and the need to analyze and react, now that you have been through this on this particular project. Are there any lessons learned for others.
If you're facing transactional issues and you haven't thought about a big-data platform as part of that solution, what do you offer to them in terms of maybe lighting a light bulb in their mind about looking for alternatives to traditional middleware.
Mohsin: Like so many people try to do, we tried to see if anyone else had done this. One of the issues in big data at least today is that you can’t find a whole slew of clients who have already gone live and who are in production.
There are lots of people in development, and some are live, but in our space, we couldn't find anyone who was live. We solved that issue via a quick-hit POC. The big lesson there was that we scoped the POC right. We didn’t want to do too much and we didn’t want to do too little. So that was a good lesson learned.
The other big thing is the data-migration question. Maybe, to some extent, this problem will never be solved. It's not so easy to pull data out of legacy database systems. Very few of them will give you good tools to migrate away from them. They all want you to stay. So we had to write our own tooling. We scoured the market for it, but we couldn’t find too many options out there.
Understand your data
So a huge lesson learned was, if you really want to do this, if you want to move to big data, get a handle on understanding your data. Make sure you have the domain experts in-house. Make sure you have the tooling in place, however rudimentary it might be, to be able to pull the data out of your existing database. Once you have it in the file system, Vertica can take it in minutes. That’s not the problem. The problem is getting it out.
We continue to grapple with that and we have made product enhancement recommendations. But in fairness to Vertica, this is really not something that Vertica can do much about, because this is more in the legacy database space.
Gardner: I've heard quite a few people say that, given the velocity with which they are seeing people move to the cloud, that obviously isn't part of their problem, as the data is already in the cloud. It's in the standardized architecture that that cloud is built around, if there is a platform-as-a-service (PaaS) capability, then getting at the data isn't so much of a problem, or am I not reading that correctly?
Mohsin: No, you're reading that correctly. The problem we have is that a lot of companies are still not in the cloud. There is still a lingering fear of the cloud. People will tell you that the cloud is not secure. If you have customer information, if you have personalized data, many organizations don't want to put it in the cloud.
Slowly, they are moving in that direction. If we were all there, I would completely agree with you, but since we still have so many on-premise deployments, we're still in a hybrid mode -- some is on-prem, some is in the cloud.
Gardner: I just bring it up because it gives yet another reason to seriously consider cloud. It’s a benefit that is actually quite powerful -- the data access and ability to do joins and bring datasets together because they're all in the same cloud.
Mohsin: I fundamentally agree with you. I fundamentally believe in the cloud and that it really should be the way to go. Going through our very recent go-live, there is no way we could have the same elasticity in an on-prem is deployment that we can have in a cloud. I can pick up the phone, call a cloud provider, and have another machine the next day. I can't do that if it’s on-premise.
Again, a simple question of moving all the assets into the cloud, at least in some organizations, will take several months, if not years.
Listen to the podcast. Find it on iTunes. Get the mobile app for iOS or Android. Read a full transcript or download a copy. Sponsor: HP Enterprise.
You may also be interested in:
- How Localytics uses big data to improve mobile app development and marketing
- Full 360 takes big data analysis cloud services to new business heights
- HP hyper-converged appliance delivers speedy VDI and apps deployment and a direct onramp to hybrid cloud
- Enterprises opting for converged infrastructure as stepping stone to hybrid cloud
- How big data technologies Hadoop and Vertica drive business results at Snagajob
- Zynga builds big data innovation culture by making analytics open to all developers
- How big data powers GameStop to gain retail advantage and deep insights into its markets
- Data-driven apps performance monitoring spurs broad business benefits for Swiss insurer and Turkish mobile carrier
- How Malaysia’s Bank Simpanan Nasional implemented a sweeping enterprise content management system
- Redcentric Uses Advanced Configuration Database to Focus Massive Merger Across Multiple Networks
- HP at Discover delivers the industry's first open, hybrid, ecosystem-wide cloud architecture
- How Tableau Software and Big Data Come Together: Strong Visualization Embedded on an Agile Analytics Engine
- Big Data Helps Conservation International Proactively Respond to Species Threat in Tropical Forests
- How Globe Testing helps startups make the leap to cloud- and mobile-first development
- GoodData analytics developers on what they look for in a big data platform