IBM’s Virtualization Alternative

Edit: Still some pretty good arguments in favor of PowerVM. Awareness is still an issue.

Originally posted November 23, 2010 on AIXchange

Did you know that when IBM publishes server benchmarks, these workloads always run on virtualized IBM Power Systems machines? The virtualization is built into the hardware and firmware; there is no concept of a non-virtualized, standalone Power machine anymore. Contrast that with offerings from other virtualization solutions running on other platforms that can degrade performance 30 percent just by using their virtualization software solutions.

The previous statements come from a recent IBM presentation. As you likely know, IBM has been at this virtualization game for a generation or so. The company developed the hypervisor that would become VM on the mainframe in 1967. In 1973, IBM was doing physical partitioning.

Here’s some more material I gleaned from this training session:

  • IBM Power Systems servers provide up to twice the performance of other virtualization solutions on other platforms. These numbers can be even greater depending on the level of virtualization you employ.
  • IBM Power Systems servers are scalable, both in terms of being capable of accommodating workload spikes and in allowing an enterprise to grow its business.
  • PowerVM technology gives you enterprise quality of service virtualization capabilities with higher performance, more scalability and enterprise security. You can have higher utilization of your machines–around 90 percent–which enables you to consolidate your workloads onto fewer physical servers. You can dynamically move from as little as 1/10th of a core to as many as 256 cores in your LPAR, using all of the resources of your server. You can make dynamic changes to resources like CPU, memory and I/O, and you can add and remove dedicated I/O adapters and storage devices, all without a reboot.
  • Live Partition Mobility allows you to easily move running workloads to other frames in your server environment. You can also use LPM to move workloads between POWER6 and POWER7 machines in your environment.
  • Using IBM Systems Director, VMs can be moved automatically to any physical machine in your environment, based on the criteria that you set up. If you have a busy workload on one machine, and more capacity available on another machine, Director can move that workload, without interruption and without human intervention, to the less busy machine.
  • IBM Power Systems servers are secure by design. No common vulnerability exposures (CVEs) have been reported against PowerVM virtualization by US CERT or by MITRE Corp. In contrast, more than 200 VMware-related vulnerabilities are listed in the U.S. government National Vulnerability Database (NVD). VMware is a third-party software add-on, while PowerVM is integrated into the server firmware. No PowerVM vulnerabilities are currently listed in the NVD. Compare PowerVM virtualization with VMware for instance.
  • POWER7 servers offer LPM, live application mobility, partition availability priority, first failure data capture, processor instruction retry, alternate processor recovery, dynamic processor deallocation, dynamic processor sparing, extended error handling and I/O adapter isolation.

The presentation featured a detailed comparison of PowerVM and VMware, making IBM’s case that PowerVM virtualization runs workloads more efficiently than VMware, with far superior resource utilization, price/performance, resilience and availability. PowerVM technology outperforms VMware by up to 65 percent on Power 750, running the same Linux workloads and virtualized resources. See this comparison of PowerVM and VMware virtualization performance for more information. In addition, PowerVM on a Power 750 will scale better than VMware with linear scaling that maximizes resource utilization with 4X more virtual CPUs. And compared to a large-tier POWER7 model such as the Power 795, you can have 32X more virtual CPUs than VMware.

Assuming I have my facts right (I borrowed them from the presentation, please correct me in Comments if you disagree with the information) VMware ESX 3.5 allows for four virtual CPUs per VM, 64 GB per VM, 192 VMs on a server, 32 CPU threads on a server and 256 GB on a server. ESX 4.0 allows for eight virtual CPUs per VM, 255 GB per VM, 320 VMs on a server, 64 threads on a server, and 1024GB on a server. PowerVM allows for 256 virtual CPUs per VM, 8192 GB memory per VM, 1000 VMs on a server, 1024 threads per server, and 8192 GB on a server.

With PowerVM technology, you can utilize all CPU cores and all physical memory. Which would you prefer for your enterprise workloads?

Let’s look at flexibility once your VM is running. PowerVM virtualization allows you to make dynamic changes to virtual CPUs, memory and I/O devices, and have integrated LPAR and WPAR support with PowerVM.

None of this is possible with ESX 3.5. With ESX 4.0, you can add but not remove virtual CPU, add but not remove memory. You can only make some dynamic I/O device changes, and limited direct access to I/O devices.

The same arguments can be made with OracleVM Server for SPARC or HP Integrity VM 4.0. Oracle/Sun allows for Sun Logical Domains on UltraSPARC T1/T2 servers only–they allow for 32 partitions on a T1 or 128 on a T2. You can add or remove CPU, but only add virtual I/O. You can perform warm migrations with constraints. There’s no support for dedicated I/O. With HP you can have 8 CPUs max, 64 GB ram. To do dynamic logical partitioning you need to reboot your LPAR. There’s no support for dedicated I/O. There’s no dynamic CPU sharing.

I still find that some shops simply aren’t aware of all that IBM Power Systems servers and PowerVM technology have to offer, and all that they can do. These customers either aren’t yet virtualizing their systems, or they don’t see the limitations they’re under using other vendors’ solutions. Hopefully comparisons like these will cause them to take a close look at IBM’s alternative.