Another 8 am start. Not so many people at this one. Guess the first night in Vegas took its toll.
The Future of Infrastructure for Cloud, Data, and Engagement
Stephen Leonard - Creating new client experiences and engagement through mobile technology
Stephen is very excited by mobile and wearables. The bulk of data generated in the mobile and social space is unstructured, and that is growing faster than structured data. Sees a shift from social/gaming/small developer to enterprise applications.
Future is more sales-focused use of your devices: location-based marketing, tracking, behavioural analysis. “The company can always be in touch with your clients.”
(That might be great is your selling, but it just creeps me the fuck out. Getting old, clearly.)
Paul Brody - IBM and Apple
Vice-President of Mobile and the Internet of Things.
Apple want IBM’s enterprise background. “We’re two companies who obsess about security and privacy.”
(Really? That’s not what I just heard…)
“Don’t treat mobile as a computer with a small screen and no keyboard.” Big pitch on security scares and pushing support for device management, rejig the front-end software portfolio to Apple-first.
Basically, if you want IBM mobile you’re getting Apple, if you want Apple in your business you’re getting IBM.
Graham Phillipsh - Traveport
Travel commerce platform - a B2B company; broker between airlines, hotels, and so on, and online agencies. Also provide processing to some airlines (e.g. Delta, Emirates). It’s a dynamic business with shifting models - which seat do you want? When do you want to board?
Mainframes are the heart of their business. They use zTPF - they believe it’s better than zOS for high-volume, low-latency workloads - and zLinux. zTPF is their system of record.
Ross Mauri - GM System Z
“The demographic of One”: personalised interaction.
Analystics is a game-changer for many companies. You should be doing them in real-time, next to and integrated with your transactional processing. Ross makes a pitch for doing that on the mainframe to avoid ETL on the grounds of expense - the idea that it’s very costly just to move the data around.
It’s interesting how many times the idea of “citizen-centric applications” has come up - in this session it was the City of Honolulu, but the same story has come from Croatia as well. I wonder how many of those engagements result in open data as well as applications, but it’s still seems like a good trend.
Big Data and Analytics 101
IBM are keen to have people review the idea that the Z is solely a transactional system. This runs counter to their own thinking from the past.
IBm are seeing CEOs start to care about technology in a way they never have before; technology has become seen as a critical factor in business sucess. This is a big change from the past where CIO/CEO priorities were quite differnt.
The biggest change is the move from finding customers for the offers we’ve predetermined to working out what the customer wants.
Much of the session is a reiteration of other sessions in this space; the key point being hamered hoome is that IBM are claiming that the tighter integration from DB2 accelerators and Hadoop on zLinux lets you do near-real-time analysis on this data.
Open Virtualization with KVM
- Initially x86_64; KVM is the hypervisor and QEMU provides the IO device emulation.
- Tony doesn’t find the question of being a Type 1 or Type 2 hypervisor particularly interesting.
- Full virt and supports any client operating system on x86; on Power it only supports Linux as a guest due to limitations in the IO virtualisation.
- KVM is now available for ARM, Power, s390x, and x86_64. Power needs to be Power8.
Why is KVM Important for customers?
SPECvirt_sc2013 is the benchmark of choice for understanding hypervisor efficiency and performance. More tiles=better performance. It is not specific to hypervisor.
When IBM and HP published results for VMWare 5.5 vs KVM on RHEL 6.5; KVM gets significantly superior numbers. IBM have not published results for PowerVM.
It’s consistent cross-platform (e.g. Fedora 19+ support KVM on s390x). It’s also very popular and heavily developed - e.g. Google and HP public clouds are KVM based.
The most important thing to understand is that KVM/QEMU are processes tha rely completely on the standard Linux facilities. If you know how to manage a Linux process, you know how to manage a KVM hypervisors. SELinux, namespaces, nice, ionice, cgroups, all apply just as well to VMs as to processes because they’re the same thing.
You can use KSM for memory de-dupe.
Three different IO architectures:
- Emulated devices; e.g. e1000 network device. Slow but don’t require paravirt drivers.
- VirtIO: VirtIO drivers are high performance.
- Device assignment: Compatible and high performance, but limits sharing and less secure. PowerKVM only has this in preview.
- Upstream for KVM on Power and KVM on System Z.
- Performance enhancements for KVM on x86_64.
- Linux scheduling - IBM have identified that transactional and batch guests mixed on the same KVM host can cause resource starvation for the transactional guests.
- IBM haven’t got any proprietary code in the core KVM area.
- IBM are working on the open source Kimchi for managing small numbers of hosts. Works on x86_64 and Power, ships with PowerKVM.
- They are also supporting PowerVC and Wave; as proprietary tools. PowerVC is more aimed at the same VMWare/RHEV niche, Wave is cloud tooling.
- OpenStack and anything that can use it.
Tony recommends Kimchi as a nice toe in the water for managing KVM. He recommends against it after you have more than a handful of hosts. IBM have a number of KVM related books: a PowerKVM Red Book, a KVM Best Practises Book (which he highly recommends), and a KVM security book.
I have to say I’m personally a little hinky about Red Books - in theory they’re written on the back of a lot of real-world experience, but I found when trying to implement J2EE apps on zLinux that the then-current zLinux Red Book had a number of suggestions that were hugely countrproductive. Hopefully that’s just an outlier.
This was a good session overall, considering it was meant to be fairly high-level, and it dovetailed nicely with subsequently seeing the Tyan Power 8 motherboard at the exhibitors’ part of the conference, running Minecraft; it’s a standard ATX form-factor with a bunch of SATA, PCIe, and DIMM slots, drops into a regular case and runs off a regular power supply. Very shiny, much geek.