VMworld 2008 Day 2 General Session

Dr. Stephen Herrod is the featured speaker of the VMworld 2008 day 2 general session. He’s the CTO & senior VP of R&D. This post will be more notes than a coherent piece. I’ll follow up with some thoughts later today.

• Goal of the infrastructure layer is to aggregate resources and be as efficient as possible.
• vCompute layer: things like FlexMigration. Focused on the VM, making it as powerful as possible. Grown from 2 vCPUs to a future 8 vCPUs, 40 Gbps, 256 GB per VM, 200000+ IOPS.
• Next generation of resource pools, up to 64 nodes in a single cluster, 4096 processor cores, 64 TB of RAM, 6 million IOPS, all running under DRS.
• Distributed Power Management continuing in development. Their VMmark exercise saved 50% of the power with DPM enabled.
• vStorage: leverage all the advanced technology of the storage in a common framework. VMFS continues forward, with enhancements to Storage VMotion. Thin provisioning will be added, and linked clones will enable better patching methods. Focus will also be on vendor management tools, enabling an API for them to interact with the VDC-OS.
• Thin provisioning is great, but also a lot of work to help with overcommittment problems. They also want to leverage hardware methods of lazy allocations, too.
• vNetwork: VMotion doesn’t preserve state between ESX servers, which is a problem for things like security. New model is the vNetwork Distributed Switch, which is a cluster-wide implementation of the virtual switch, and VMs will take their network profiles with them when they VMotion. First third-party switch is the Cisco Nexus 1000V, which is nice because network admins can interact with the virtual switches like any other Cisco equipment.
• Application vServices layer: everything will work for existing application loads, as well as new ones that we don’t know about yet.
• vApp: existing VMs can be wrapped in the vApp container. Based on OVF, encapsulates all the different VMs that make up an application, also represents SLA information which represents a contract with your VDC-OS.
• A variety of technologies to help with planned and unplanned downtime, like NIC & HBA teaming, VCB, HA, VMotion, Storage VMotion, and at a high level, Site Recovery Manager.
• Fault Tolerance: focused on unplanned downtime for an application. Idea is that a shadow copy of a VM is running somewhere else, so if something happens the other VM will be available. A demo of the FT technology by Mark Vaughn of First American Corp. FT is simple to enable with a right-click in the new vCenter client. It basically starts a VMotion to copy the running state over, and then enables vLockstep to mirror the activities of the primary VM. FT helps with a lot of human error, cords coming out, powering off the wrong box. When something happens and FT kicks in it will automatically start reprotecting itself within the same resource pool.
• Security vServices: be able to describe the services in the vApp definition. The building block for the security services is a set of APIs called VMSafe. Security software is working outside the VMs themselves. Relies heavily on the distributed network switch. This means new VMs are automatically protected. It’s “right-sized” security capacity, which grows easily with your environment.
• Management vServices: VirtualCenter becomes vCenter, and way more extensible. Partners can plug in for extensibility and also for “pane of glass” sorts of things. ConfigControl, Orchestrator, CapacityIQ, and Chargeback all new modules to do cool things with the infrastructure.
• AppSpeed helps monitor applications and work with the VDC-OS to take actions. It discovers the topology of services in a non-intrusive way. Then looks at QoS from an end-user perspective, and can deduce when things are wrong. It’s a totally new way to collect information in the virtual framework. A demo from Asaf Wexler, formerly of Beehive (now with VMware). As the application is used a map of the dependencies is created automatically. The demo vCenter plugin was built overnight by the AppSpeed team. Stats available can help identify root causes of problems. Policy engine can take actions based on stats gathered, to fix problems automatically.
• vCenter: lots of work on alarm capabilities. Getting ported to Linux, also to be shipped as a virtual appliance. Adding multi-platform client technology to run on Macs, Linux, iPhones, etc. On the iPhone, if you shake it you trigger a disaster recovery scenario (joke!). 🙂
• Cloud vServices: “cloud” most abused phrase since “virtualization.” A model to quickly use resources, to pay for only what you need, accessible with standard protocols, scalable & elastic, and economies of scale. “Local cloud” exists in your own data center. Needs to be more flexible. App compatibility maintained into the cloud, without having to rewrite an app. Standardization removes complexity in apps and lowers switching costs. Last, provides multi-tenancy in a resource pool, where others can come in and be assured their security and performance needs are met.
• vCloud APIs: image management, user accounts, chargeback, mobility, all APIs needed to bridge the gap between local clouds and off-premise clouds.  Clouds are federated, and the vApp/OVF definitions help determine where an app could run, whether it can be outside a firewall, for example, or it needs to stay internal.
• VMware View: targets applications and data, following a user and not their hardware. Use mobile devices, home computers, etc. securely and with proper management. View evolves VDI, User experience changes based on where you are, whether on a WAN, LAN (richer experience), or local (with graphics cards, etc.).  New technologies to cache VMs locally. Unity in Fusion and ACE contribute to these new technologies. A very thin layer to avoid other OS software. Demo by Jerry Chen of VMware, demonstrating setup tasks using View Manager and View Composer. Linked clones saves a lot of disk space while speeding the provisioning time. New client hypervisor runs VMs locally on hardware but lets you manage the VM from View Manager. Demo of adding Google Chrome to all the desktops using ThinApp, update master image, then notifies clients to restart. Jerry Chen then rickrolled the attendees to prove that Chrome works on the clients. Thanks. 🙂

In the end, irtualization is proving to be a broad problem-solver. 🙂