Edit: There have been a few updates since I first posted this.
Originally posted October 12, 2010 on AIXchange
I first heard about an updated version of the virtual I/O server (VIOS) during a recent IBM conference call. Now it’s official.
We already use VIOS for sharing disks and networks, active memory sharing and live partition mobility. With these just-announced enhancements, we’ll be able to suspend and resume workloads, do more with virtual networks and take advantage of thin storage provisioning and storage pool sharing capabilities.
Here are some announcement highlights, starting with a new feature called suspend/resume.
As I learned in the conference call, suspend/resume is the process of “freezing” an LPAR and saving the complete system state to disk. Then you can restart the workload exactly where it left off, without data loss. The entire LPAR system state is stored in a set of files and can be resumed on either the same server or a different system after migration. After suspension, the server resources are freed up for use by other workloads.
As you can imagine, this feature can make hardware maintenance much easier, because it allows system administrators to perform system updates or CEC upgrades without the need to shut down and restart applications, and without the need to engage application teams to verify that everything is running properly after the restart.
Where live partition mobility allows us to shift resources between physical machines while applications are still running, with suspend/resume, we’ll be able to move workloads to another machine (though obviously with an interruption of services). We’ll also be able to temporarily suspend low-priority or long-running workloads to allow more urgent processes to access server resources.
For debugging or forensics purposes, IBM states that a workload can be temporarily suspended and a copy made for offline analysis for security or performance purposes. I can’t wait to test out this intriguing feature.
More about suspend/resume from the IBM announcement letter:
“Using Suspend/Resume, clients can provide long-term suspension (greater than 5-10 seconds) of partitions, saving partition state (memory, NVRAM and VSP state) on persistent storage, freeing server resources that were in use by that partition, restoring partition state to server resources, and resuming operation of that partition and its applications either on the same server or on a different server.
“Requirements for Suspend/Resume: All resources must be virtualized prior to suspending a partition. If the partition is to be resumed on a different server, then the shared external I/O (disk and LAN) should remain identical. Suspend/Resume works with AIX and Linux workloads when managed by HMC.”
Here’s what the announcement letter says regarding shared storage pools, VIOS grouping and thin provisioning:
“VIOS 2.2 allows the creation of storage pools that can be accessed by VIOS partitions deployed across multiple Power Systems servers so that an assigned allocation of storage capacity can be efficiently managed and shared. … Multiple VIOS 2.2 partitions can utilize a common shared storage pool to more efficiently utilize limited storage resources and simplify the management and integration of storage subsystems.”
During the conference call the presenters mentioned that this would eliminate the need for vscsi devices or NPIV, but I’ll need to do some hands-on testing to understand the functionality better.
“VIOS 2.2 supports highly efficient storage provisioning, whereby virtualized workloads in VMs can have storage resources from a shared storage pool dynamically added or released as required.”
It sounds like the thin provisioning that we’re used to managing on our storage subsystems can now be managed from our VIO servers. I look forward to testing it out.
“When a new VM is created, the amount of physical storage used is less than the amount defined for the virtual workload, resulting in optimal storage utilization across the shared storage pool. Additional
storage is delivered dynamically when workloads expand and released when workloads contract. This automates optimized storage utilization, has a more cost-efficient use of storage resources and integrates multiple storage subsystems.”
The last thing they touched on in the training was the enhancements to the virtual networking.
“The virtualized network switch functionality within the VIOS will include support for SNMP, networking QoS, dynamic VLAN and MAC access control lists (ACLs). There will be more sophisticated controls for monitoring and tuning network traffic between virtualized workloads. There will be control over networking QoS (quality of service) rules for specific LPARs and you can fine-tune the performance of network-sensitive workloads. There will be support for MAC based access ACLs to allow administrators to impose higher levels of protection for specific workloads.”
According to the announcement letter, VIOS 2.2 is set for availability on Oct. 15.