AI needs the right information architecture

Source: Vsoft, 26/11/2018

`We say: take the analytics to the data, don`t take the data to
the analytics. Copying and extracting data allows it to lose value
and context, and causes latency, which makes the data out of
date,` says Karmjeet Kahlon, VP, Worldwide z Hybrid Cloud, IBM.
At the IBM `Unleash the Heart of your Enterprise` executive forum
in Johannesburg this week, IBM shed light on solutions designed to
support next-generation analytics in the enterprise.
On the theme `there`s no AI without the right IA`, Tom Ramey,
director, z Analytics at IBM, said challenges facing enterprises
include modernisation of the mainframe, cost, complexity and
security, and crucially, leveraging analytics and machine-learning
in real-time.
`50% of S&P 500 companies are being replaced every 10 years. The
lesson here is you have to continue disrupting and growing your
revenue. Machine-learning and artificial intelligence (AI) can be
tools to support this disruption.`
Ramey said leading CIOs are well aware of the potential for
machine-learning and AI. `81% of CIOs believe AI to be very
important or extremely important to the future of their
organisations.`
Analytics challenges have become overwhelmingly complex for many
organisations, said Ramey. Data often traverses a complicated
journey, involving multiple copies of the data moved off the
database.
He said the average mainframe customer moves 1TB of data per day
off the platform to somewhere else for analytics at a cost of $10
million over four years.
`Moving data for analytics increases cost, puts the data at risk
due to many copies being made, and creates a latency gap, when our
goal now should be to get down to real-time for analytics. We need
our analytics systems to be just as resilient and available as our
transaction systems nowadays.`
The data gravity approach, to perform analytics where the
preponderance of the data originates, or hybrid transaction and
analytical processing, marrying transactions with analytics, has
emerged as the way to deliver faster, simpler, more cost-effective
analytics in real-time.
`We came up with the Db2 Analytics Accelerator and plugged it in
to the mainframe, with the Optimizer deciding the best path to
process the query, to deliver [up to] 3 000 times faster query
processing that is also super simple, highly secure and saves tons
of money.
`Customers are now running reports that took eight hours, in under
a second. I would argue that we`ve evolved so far beyond just
taking existing queries and speeding them up. Customers have
figured out it`s not just about the queries you run today, it`s
about all the queries you can`t run today.`
The mainframe is a data-serving juggernaut, and Db2 is all about
supporting enterprise-scale next-generation applications on the
mainframe, he said. Db2 V12 delivers concurrent queries, up to 100
times faster, to provide deeper insights, with enhanced support
for cloud and mobile workloads and a 23% lower CPU cost by
providing in-memory techniques as well as continuous availability,
scalability and security, states IBM.
Ramey said: `Machine-learning is the hot new ticket these days. It
identifies historical patterns in data on the mainframe,
identifies patterns in that data, and uses statistical algorithms
to build a model based on those algorithms. We learn from that
data and then we can take that model and we can provide actionable
insights by using it against the new data that comes in. We use it
to make predictions and recommendations.
`With machine-learning, companies can truly tap into the rich vein
of data in their historical system of records. This means instead
of only using analytics to run reports on data from the past, we
can now predict what will happen in the future.`
But there are challenges and as little as 5% of commercial data
science projects make it into production, he said.
`Traditional machine-learning processes require significant human
intervention, take months to get the data, prepare the data, train
a model and finally deploy a model. So we came up with machine-
learning for z/OS that dramatically simplifies that long process,
utilising data gravity and helping the data scientists.`
`Data gravity is a critical concept now,` said Martin Blignaut,
enterprise and mainframe software sales leader, IBM South Africa.
`You need to move the analytics to the data, not the data to the
analytics. It`s powerful, as it defines where we are going and how
we see the management of information on this platform in future.`