Availability is a crucial attribute for modern servers. Organizations routinely operate 24/7, and planned uptime can reach four, five, and even six "9s". In non-virtualized server environments, the sheer number of physical servers limits the affect of faults -- if a single server fails, only the application on that server becomes unavailable.
The risk is exacerbated in a virtual setting because a physical server fault can take down all of the VMs running on that server. Consequently, data center professionals must take extra care to ensure the proper ongoing operation of physical server hardware.
Resource shortages are one of the biggest threats to virtual server availability. Not only will inadequate CPU, memory, disk I/O or network bandwidth impair performance, it can also cause VMs on the server to crash and sometimes take down the entire physical server. Hardware upgrades can be an effective way to fend off resource shortages and prevent server problems.
Software monitoring tools can help you determine the actual resources available in (and used by) a physical server, There are numerous third-party monitoring tools available such as Vizioncore's vFoglight or HP OpenView, along with open source tools like GroundWork monitoring software.
Still, experts note the value of monitoring and management tools provided by major hypervisors from Microsoft, VMware or Citrix. "Because they're giving away the hypervisor, they need to make money off of those management tools, so they've invested pretty heavily in them," said Dave Sobel, CEO of Evolve Technologies, a solution provider headquartered in Virginia.
Once a tool is selected, it can be used to examine a large range of performance counters, capture and log data over time, send alerts when excess utilization occurs and help to identify utilization trends that data center administrators can use to plan server resource upgrades.
Maintain availability by ensuring adequate resource headroom in each virtual server. For example, if an organization has two servers and each server hosts four VMs, there should be enough available resources in either server to potentially host all eight virtual machines if the companion server fails. With more virtual servers available in the data center, that load is easily spread out over more machines. As an example, if there are eight virtual servers and each runs six VMs, each of the remaining servers should have enough resource headroom to support at least one VM from the failed machine. Remember, headroom should be included in all server resource requirements, and upgrades may be appropriate to achieve the necessary resource levels.
One way to manage the expense of hardware upgrades is to approach virtualization in phases. "Don't do so much all at once that you get overwhelmed," Plankers said. "Give it time to be able to adjust and learn from the workloads -- especially if you're upgrading your environment and virtualizing old servers onto newer hardware."
Of course, planning for such failover behaviors (that is, selecting which VM goes where in the event of a fault) is an important aspect of the availability puzzle. This is an underlying idea of load balancing in a virtual server environment.
Virtualization platforms can certainly make failover decisions automatically and restart crashed VMs on any other server where adequate resources are available. But there are many cases when certain VMs should be kept apart or failed over only to specific servers. Failover planning should include the development of rules that tools such as VMware's DRS (Dynamic Resource Scheduler) can use to move virtual machines.
ABOUT THE AUTHOR
Stephen J. Bigelow, senior features writer, has more than 15 years of technical writing experience in the PC/technology industry. He holds a BSEE, CompTIA A+, Network+, Security+ and Server+ certifications and has written hundreds of articles and more than 15 feature books on computer troubleshooting, including Bigelow's PC Hardware Desk Reference and Bigelow's PC Hardware Annoyances. Contact him at firstname.lastname@example.org.