ibminterconnect

generally the trend toward in-memory is real and taking hold

look at Hadoop and Spark

also seeing other in-memory approaches catching fire (pun intended) in the #BigData world - Apache Spark is all anybody was talking about (mostly) at #BigDataSV and Strata + Hadoop World

. @jeffreyfkelly to me the open source moment around in-memory and how it ties to analytics then to the apps #developers

ever since there's been memory and databases there's been talk of in-memory and with prices dropping the way they have and flash coming to the fore it makes more and more sense

There are huge benefits but they come at a cost -- memory is expensive!

i think in-memory is getting so much attention because of what it enables - analytics at a speed that allows for you to make and execute a decision while it still has an impact

Flash technology is doing a lot to change that equation for the better

flash attacks the latency from "the horrible storage stack"

@anirbahn true, cost came down dramatically but its still not "cheap" - seeing a tiered approach with disk, flash and DRAM depending on "temp" of the data in questions

@dfloyer and I wrote about IBM's approach not long ago ... http://premium.wikib...

actually I think flash is becoming cheap - I'm going to post a video with Eric Herzog that discusses this

#PowerSystems has a feature called CAPI that speeds up flash latency to be much closer to DRAM. Great for #nosql apps for example

The ability to process and analyze all that big data is paramount, without the right infrastructure all the data can be useless #IBMInterConnect

It's not just ability, it's speed too, right? Really transforms how you can use the data if the answers come back quickly.

yes, need to balance deep data analytics with inline analytics (aka operational analytics)

requires different storage and compute infrastructure depending on use case (real-time requirements)

. @jeffreyfkelly decoupling the infra from the software at the app layer is very important he talked about it on @theCUBE yesterday http://premium.wikib...

as a customer I want to reduce complexity. Save on people time

@dvellante yes, reduce complexity whenever possible, but show;dint do so at the expense of the desired outcome

in other words, it would be ideal if there was one infrastructure to support ALL analytics (offline and operational) workloads, but in reality you'll probably need more than one approach

but definitely move towards less complexity and bringing together transactions and analytics

Do you think using tools like virtualization & cloud orchestration can get you closer to that "one infrastructure" dream?

@anirbahn yes - tying together hybrid cloud environments with common tools and workflows

a must watch

the key point he kept hammering was that flash was now less expensive than so-called hi spin speed disk

Hi performance disk is an oxymoron - it doesn't exist. It's like saying that's a fast turtle

@dfloyer predicted in 2009 that by 2014 high performance SaaS and FC disks would be killed by flash - maybe he was a few months off but that was a great prediction

With data reduction techniques you can dramatically reduce flash costs and b/c of the speed you don't pay nearly the perf penalty as you would w/ disk

for me, in-memory allows us to take the business conversation from a theoretical one (i.e. what could I do if I could act in time to save the customer? prevent the fraud? etc.) to a reality.

Consider the savings that flash provides with regards to speed, density and power efficiency

The point is with flash now attacking disk latency it opens the door for a balanced systems approach with more horsepower at the system level - flash is forcing the pendulum to swing back to server performance

a few years ago, Steve Mills declared "Mark my words, Sparc is dead" - others said the same about Power

I'd love to hear what the crowd thinks about this one!

What's happened is #bigdata workloads are a tailwind for systems that have affinity in historic HPC workloads

HPC meets Big Data is the theme there and it's real imho

I think the market interpreted IBM's move to sell off its commodity HW biz to mean it was out of the #BigData hardware business. Not so, IBM pushing hard on Power to improve efficiency and performance in #BigData analytics enviroments

If you look at the computation patterns between #bigdata and traditional #HPC, they have a lot in common. #HPC has a bright future.

@anirbahn Fair to say #BigData makes #HPC accessible to, affordable for the enterprise?

#BigData has certainly driven the need for #HPC capabilities with enterprises (and even smaller businesses)

Open source is a powerful wild card - it's like making Joker's Wild in a card game - it's there and can change the game

agreed - I expect acquisitions in the #BigData space, but open source nature of the community ensures innovation continues

Opening up Power tries to capitalize on the ARM playbook applied to the enterprise in a way

It can transform and accelerate how businesses can create and deploy new tools

100 partners has real juice - they're not just "Barney Deals"

you never know where the next innovator will come from

Making IBM OpenPower is a smart move by IBM - big demand from cloud service providers for multiple sourcing

There are some well known partners there too - Google, NVIDIA, Samsung, Redis, ...

@anirbahn It allows partners to intergrate and innovate following an open set of rules, not the rules of a single platform provider

Basically it's data compatibility. Most x86 reads/writes data in little endian (byte ordering). Power can now interact natively with that data.

@anirbahn the implications meaning better performance? Lower latency?

It means more application portability and support for clients who wish to leverage both Linux, open source tools and Power systems

The ability to use Power gets you better performance, yes. Faster results for your queries.

Power has 4 times the threads as x86, tons of memory bandwidth, CAPI acceleration, etc.

@DouglasGibbsIBM that's key - #BigData requires flexibility, agility; application portability an important capability