My perspective on the Teradata Aster Big Analytics Appliance

Aster Big Analytics Appliance 3H

By now, no doubt you have heard the announcement of our new Teradata Aster Big Analytics Appliance. In the interests of full disclosure, I work for Teradata within the Aster CoE in Europe. Prior to joining Teradata,  I was responsible for a large complex Aster environment which was built on commodity servers in excess of 30 TB of usable data with a 24 x 7 style operational environment. So my perspective in this post is from that standpoint and also recalling the time when we went under a hardware refresh and software upgrade.

OK, First of all you procure 30 servers and at least two network switches (for redundancy). When you receive them, it up to your data centre team to rack them and cable them up. Next,  check the firmware on each system is the same, surprise, surprise they aren't, so a round of upgrades later, then you configure the raid controllers. In this case we went for Raid 0 which maximises space, more on that choice later...

Then it is over to the network team, to configure the switches and the VLAN we are going to use. We then put on a basic Linux image on the servers so we can carry out some burn in tests, to make sure all the servers have a similar performance profile. Useful tests, as we found two servers whose raid controllers were not configured correctly. It was the result of human error, I guess manually doing 30 servers can get boring. This burns through a week, before we can install Aster, configure the cluster and bring all the nodes online to start the data migration process. Agreed, this is a one off cost, but in this environment, we are responsible for all hardware issues, network issues, Linux issues, with the Vendor just supporting Aster. Many customers never count that cost or possible outages that might be avoided because of these one off configurations.

We had some basic system management as these are commodity servers but nothing as sophisticated as Teradata Server Management and the Teradata Vital Infrastructure. I like that with the Server management software it allows me to manage 3 clusters within the Rack logically (e.g. Test, Production, Backup). I also like the proactive monitoring, as it is likely they will identify issues prior to them becoming an outage for us, or an issue found with one customer can be checked against all customers. If you build and manage your own environment, you don't get that benefit.

Your next consideration should be when looking at an appliance, is it leading edge and how much thought has gone into the configuration? The Appliance 3H is a major step forward from Appliance 2. From a CPU perspective, it has the very latest processors from Intel, dual 8 core Sandy Bridge @ 2.6GHz. Memory has increased to 256GB. Connectivity between nodes is now provided by Infiniband at 40Gb/s. Disk drives are the newer 2.5 size, enabling more capacity per node. Worker nodes using 900GB, while backup and the Hadoop nodes leveraging larger 3TB drives. RAID 5 for Aster and RAID 6 for the backup and Hadoop nodes. Also with the larger cabinet size, enables better data centre utilisation with the higher density that is possible.

I also like the idea of providing integrated Backup nodes as well, previously Aster just had the parallel backup software only, you had to procure your own hardware and manage it. We also know that all of these components have been tested together, so I am benefiting from their extensive testing, rather than building and testing reference configurations myself.

What this tells me, is that Teradata, can brings advances in hardware quickly to the marketplace. Infiniband will make make an important difference. For example, for very large dimension tables, that I have decided against replicating, joins will run much faster. Also I expect positive impact on Backups. In my previous environment, it took us about 8 hours for a full backup of 30 TB or so. Certainly the parallel nature of their backup software could soak-up all the bandwidth on a 10GB connection, so we had to throttle it back.  On the RAID choices, I absolutely concur with the RAID 5 choice. If I was building my own 30 node cluster again I wouldn't have it any other way. While the replication capabilities in Aster protects me against at least any single node failure, a disk failure, will bring that node out of the cluster, until the disk is replaced and the node is rebuilt and brought back online. When you have 30+ servers each with 8 drives (240+ disk drives) the most common failure will be the disk drive. With RAID 5, you can replace the drive, without any impact on the cluster at all, and you still have the replication capabilities to protect yourself from multiple failures.

I also like the option of being able to have an Hadoop cluster tightly integrated as part of my configuration. For example if I have to store a lot of 'grey data' e.g. log/audit files etc for compliance reasons,  I can leverage a lower cost of storage and still do batch transformations and analysis as required. Bring a working set of data (last year for example) for deeper analytics. With the transparent capabilities of SQL-H, I can extend those analytics into my Hadoop environment as required.

Of course purchasing an appliance, is a more expensive route than procuring it, building and configuring it all yourself. However, most enterprise are not hobbyists, and building this sort of infrastructure, is not their core competence nor is is bringing value to their business. They should be focused on the Time to Value and with the Teradata Aster Big Analytics appliance the time to value will be quick, as everything, is prebuilt, configured, tested and ready to go, to accept data and start performing analytics on it.  As I talk to customers across Europe this message is being well received when you talk through the details.

I'll leave you with this thought, one aspect of big data that I don't hear enough of is Value. To me, the dimension of Volume, Variety, Velocity and Complexity are not very interesting if you are not providing value by means of actionable insights. I believe every enterprise customer needs a discovery platform capable of executing the analytics that can provide them an important competitive edge over their competition. This platform should have the capability to handle structured as well as multi-structured data. It should provide a choice of analytics, whether they be SQL based, MapReduce or statistical functions. It should provide a host of prebuilt functions to enable rapid progress. It should be a platform that can appeal to power users in the business, by having a SQL interface and that will work with their existing visualisation tools to our most sophisticated data scientists, by providing them a rich environment to develop their own custom functions as necessary, while enabling them to benefit both from the power of SQL and Map Reduce to build out these new capabilities. In summary that is why I am so excited to be talking with customers and prospects about Teradata Aster Big Analytics Appliance.

Don't be left in the dark... Exciting Teradata Aster Announcement coming soon

We’re gearing up for a big Teradata Aster announcement. Mark your calendars to get the scoop on  Oct 17th. We will be hosting a live Webinar to announce it. Register here for it. 

Believe me, it is going to be amazing...

The Big Data Analytics Landscape

As a technologist and evangelist working in the big data marketplace it is certainly exciting. I am excited by the new products we are bringing to market and how this new functionality really helps to bridge the gap for Enterprises adoption. It is also surreal, in terms of the number of blog posts, tweets on Big Data and there seems to be a new big data conference cropping up on a weekly basis across Europe :-)

It is interesting to monitor other vendors in the marketplace and how they position their offerings. There is certainly a lot of clever marketing going on (that I believe in time will show a lack of substance) and some innovation too, . You know who you are.... But jump aboard the bandwagon. Just because you might have Hadoop and your database + Analytics within the same rack that doesn't mean they are integrated.

It is also interesting the fervor that people bring when discussing open source products. Those people who know me, know that I am a long time UNIX guru over 20 years from the early BSD distributions to working with UNIX on mainframes, to even getting minix to work on PC's before Linux came along. Fun times, but I was young free and single and enjoyed the technical challenge. I was also working in a research department in a university. However the argument that Hadoop is free and easy to implement and will one day replace data warehousing, doesn't ring true for me. Certainly it is true is has a place, and does provide value, but it doesn't come at no cost. Certainly Hortonworks and Cloudera provide distributions that are reducing the installation/configuration and management effort, but you have multiple distributions, starting to go in different directions? MapR for example?

How many enterprises really want to get that involved in running and maintaining this infrastructure. Surely they should be focused on identify new insights that provides business benefits or gives greater competitive advantage. IT has an important role to play, but it will be the business users ultimately that need to leverage the platform to gain these insights.

It is no use getting insights, if you don't take action on them either.

Insight gained from big data analytics should be fed into existing EDW (if they exist) so they can enhance what you already have and the EDW provides you with a better means  of operationalizing the results.

I say to those people who think Hive is a replacement for SQL, not yet it ain't, it doesn't provide the completeness or performance that a pure SQL engine can provide. You don't replace 30+ years of R&D that quickly...

To the NoSQL folks, this debate is taking on religious fervour at times, It has a role, but I don't see it replacing the relational database overnight either.

In a previous role I managed a complex DB Environment that included a Big Data platform for a company that operated in the online gaming marketplace in a very much 24 X 7 environment, with limited downtime. It was the bleeding edge at times, growing very fast.  If we had Teradata Aster 5.0 then, my life would have been so much easier. Se had an earlier release but we learned a lot. We proved the value of SQL combined with the Map Reduce programming paradigm. We saw the ease of scaling and reliability, We delivered important insights into various types of fraud, and took action on them, which yielded positive kudos for the company and increased player trust, which is very important in an online marketplace. We also were able to leverage the platform for an novel ODS requirement and had both executing simultaneously along with various ad-hoc queries. I was also lucky then and since to meet real visionaries, like Mayank and Tasso which gives you confidence in the approach and the future direction

When you think of big data analytics, it just not just about multi structure data or new data sources. Using SQL/MR for example may be the most performant way to yield new insights from existing relational data. Also consider what 'grey data' already exists within your organisations, it maybe easier to tap into that first, before sourcing new data feeds. The potential business value should drive that decision though.

Do not under estimate the important of having a discovery platform as you tackle these new Big Data Challenges. Yes, you will probably need new people or even better, train existing analysts to take on these new skills and grow your own data scientists. The ease of this approach, will be in how feature rich your discovery platform is, How many built in and useful analytical functions are provided to get you started, before you may have to develop specific ones of your own.

I suppose,  some would say I am rambling with these comments and not expressing them very elegantly, but help is at hand :-). We recently put together a short webinar, I think it is about 20 minutes duration. 

The Big Data Analytics Landscape: Trends, Innovations and New Business Value, featuring Gartner Research Vice President Merv Adrian and Teradata Aster Co-President Tasso Argyros.  In the video, Merv and Tasso, answer these questions and more, including how organizations can find the right solution - to make smarter decisions, take calculated risks, and gain deeper insights than their industry peers.

• How do you cost-effectively harness and analyze new big data sources?
• How does the role of a data scientist differ from other analytic professionals?  
• What skills does the data scientist need?
• What are the differences between Hadoop, MapReduce, and a Data Discovery Platform?
• How are these new sources of big data and analytic techniques and technology helping organizations find new truths and business opportunities?

I suggest if you have the time to spare... watch the video

What do you think?

For me it is all about the analytics and the new insights that can be gained and acted upon

First Steps in Exploring Social Media Analytics

As I talk with customers and colleagues the topic of social media analytics is often featured.  Some customers have already got a strategy defined and are executing to a plan while others are at a more nascent stage but believe in the potential to have a direct and almost immediate connection to their customers.

I'll admit that I am somewhat of a social media novice so it will be a learning experience for me too. I am intrigued by the depth of analytics that may be possible. Only this month did I setup my profile on Facebook and I'm 47! I have been using Twitter more regularly of late, since our Teradata Universe conference and I probably look at it 3 or 4 times a day, depending on what's on my schedule for the day. I am finding some interesting, funny and informative updates each day, as I slowly expand the number of people I follow. It is really a mixture of friends and work related contacts at the moment.  I have been a member of LinkedIn for a number of years and find it a useful resource from a professional perspective.  I am within the first 1% of members who subscribed to this site (I received an email to that effect, that I was within the first 1 million sign ups when the site hit 100 million). Finally I am keen on blogging more frequently when I have something interesting to share (i.e. this! :-) ) I had stopped blogging for about 5 years at one point.  I have also started with Flickr and YouTube as well. I'll be my own guinea pig in some ways as I explore and experiment on possible useful analytics in these social media channels.

However when most people think of Social Media and associated analytics, Facebook and Twitter are often mentioned first.   These social media systems do provide an API that can be used to readily access data, and these split into two broad categories that reflect the social media’s attitude to customer data. 

The first approach is the open approach adopted by Twitter.  Users on Twitter are warned that their posts are visible to anyone (who can find them). The second approach is that adopted by Facebook. There is an extensive privacy model, and data needs to be accessed using authorizations (from likes and games).

Twitter

• Data is free but there is a lot of it
• Identifying relevant stuff isn’t easy
• History isn’t usually available

Facebook

• Public data is free through search
• Basic data available through application
• More sophisticated data available with permissions (using Oauth 2.0)

 

Focusing on Facebook and Twitter you see two very different levels of information.  Twitter provides only basic, row level data. Facebook provides much more complex, relational data. We'll explore these in more detail in future posts.

Data from social media must be linked in three ways:

·      Within the social media itself

·      Across multiple social media

·      Between social media and the bank’s core data

The most secure forms of linking are to use unique references: email addresses, IP addresses and telephone numbers.  This can be supported by direct access methods (i.e. asking the user for their Twitter name, or persuading them to Like the bank on Facebook from within a known environment).

However, even then the confidence in the link must be evaluated and recorded: this information is user provided and may be wrong in some cases. The notion of a “soft match” should be adopted – we think that this is the same person, but we cannot be sure.

I would like to end this post with a recommendation to read the following white paper by John Lovett from Web Analytics Demystified  Beyond Surface-Level Social Media. Lovett, who has written a book on Social Analytics , lays out a compelling vision for Deeper Social Analytics for companies.  He clearly presents the value for companies to go beyond surface level analytics of likes, followers and friends and challenges you to ask deeper and more important questions. This white paper has been sponsored by Teradata Aster and is available for free from here.

In reading this white paper you will gain an understanding of the term 'Surface-Level Social Media' coined by John and how it is possible to gain competitive advantage even operating at this level. He will outline how Generation-Next Marketing is being powered by Social Analytics backed up with a number of interesting customer examples. He goes on to outline a 7 point strategy to build your deeper social media strategy. Finally John concludes with how unstructured data can yield valuable customer intelligence.

I found it to be very informative and well written and gave me a number of new insights and points to ponder. I would be interested in your thoughts on it too. 

Enjoy!