Boot storms and how to avoid them

Boot storms and how to avoid them

Summary: Are you in the midst of a user mutiny over your VDI boot storm? There's a way out and you don't need a concrete umbrella.

SHARE:

A "boot storm" is the tremendous performance drain experienced when hundreds or thousands of users logon at, or near, the same time in VDI environments. The drag on network throughput, host performance and SAN I/O can be devastating to an unprepared VDI installation. Some of the pressure can be eased by building the environment according to some best practices laid out by virtualization vendors. But, there are a few areas that you might not have considered when constructing your VDI infrastructure. Now, you should consider them, regardless of where you are in your VDI implementation.

In our excitement to toss desktop management to the data center, it's possible that we've made a few mistakes along the way. We didn't expect boot storms to have quite the effect that they have on resources and productivity. It's significant and it's frustrating.

There used to be an old joke around offices that, in the morning, you turn on your computer, go get a cup of coffee, come back and you might see a login prompt. We thought our new, super fast technology would alleviate the need to head for the break room. It hasn't.

There are some solutions to the boot storm dilemma and you can implement them at any stage of your deployment or even in the post-deployment phase.

  • Disable Memory Ballooning - Memory ballooning sounds good in theory but for VDI, it has the opposite from desired effect. When a large number of desktops become active, hosts can run out of RAM, causing paging, *Motioning and very high disk I/O.
  • Use Single vCPU VMs - Most user applications are single-threaded and won't benefit from multiple vCPUs.
  • Timed Boots - You can manage boot storms by having virtual desktops power on in stages prior to user logon. Having the VMs ready for the user also saves time.
  • Optimize Antivirus Scanning - Antivirus software is a necessary evil but can put a drain on resources if not properly configured. Edit settings to Scan on Write, Scan Local Disks Only and exclude scanning the Pagefile.sys file and the Print Spool directory. Additionally, you can make optimizations based on your environment and special conditions.
  • Disable 3D Screen Savers - 3D Screen Savers should be removed from user's desktops or disabled. These screen savers, when engaged, consume CPU power unnecessarily.
  • Physical Domain Controllers - Yes, you can run virtual domain controllers but Microsoft still recommends using physical systems for domain controllers.
  • NIC Teaming (Aggregation) - NICs should be teamed for greater throughput.
  • Application Virtualization - Applications should be virtualized as well as operating systems. This simplifies and shrinks "Gold" images to a very supportable size and level of complexity making them easier to deploy and to update.
  • Accurate Network Bandwidth Calculations - If you're still using the old 20Kbps/user calculation, your network will suffer bandwidth latency and a poor user experience. Your estimates should take into account the types of activities that your users in engage in. A more accurate estimate (average) is in the 100Kbps/user range.
  • Proper Storage Configuration - 15K RPM SAS disks or SSDs and RAID configuration choices will determine how well your storage performs in your VDI environment. Experience shows that RAID 10 is a wise choice for VDI because of its speed. VDI environments are generally 70% write/30% read and RAID 10's I/O throughput is necessary for a good user experience. SSDs with their 4,000 to 5,000 IOPS also greatly reduce boot times but at considerable cost. The more affordable 15K RPM SAS disks comparatively lope along at 150 to 180 IOPS.

This list is by no means exhaustive but it will give you a good start. You might see two or three areas where you can make marked improvements by simple reconfiguration and without a major cash outlay.

Although I've written negatively about VDI in the past, I think that SSDs and GigE networking makes it a pretty good solution for companies spending millions in desktop support, especially when coupled with hosted applications.

Boot storms are a product of poorly designed or under-designed VDI implementations. It's a shame too, since VDI isn't cheap. In fact, in up front costs, it's much more expensive than traditional heavy desktop computing. However, it doesn't have to be the painful trip to the coffee pot that it once was. VDI, correctly configured, can be a very pleasant experience and quite economical in the long run. It can also save you from user mutiny and a very different kind of boot storm.

Related Stories:

Enterprise-level BYOD and the quest for freedom: York County, Virginia

The answer to the BYOD question is Virtualization.

Yes, desktop computing has failed but there's a solution

The numbers are in on the move to virtualization

Why desktop computing failed

In the vPresence of a Virtual Desktop

Topics: Networking, CXO, Emerging Tech, Hardware, Storage, IT Employment

About

Kenneth 'Ken' Hess is a full-time Windows and Linux system administrator with 20 years of experience with Mac, Linux, UNIX, and Windows systems in large multi-data center environments.

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.

Talkback

3 comments
Log in or register to join the discussion
  • Shades of old Mainframe VDTs!

    Reminds me of the 70s when a simple 1 page text BOM report was delivered 1/2 a day later off an IBM mainframe!
    "Back to the future?"
    ;-)
    kd5auq
  • VDI + App Virtualization = Expensive

    If you end up paying for VDI AND APP Virtualization, you are starting to talk about a very expensive solution. Probably better to seriously consider better managing indivuals PCs and then allowing users to simply Remote Control those individual PCs if they are away from the office.
    jkohut
  • Clear Skies and Tropical Drinks with IDV

    Imagine a straw sticking out of a computer monitor, and being able to enjoy a cool tropical drink from it whenever you wanted. Then one stormy day IT comes into your office and removes the desktop, replaces it with a thin client, and reconnects the straw. You ask, "how am I going to get my tropical drink?" IT replies, "The same way, just suck from the straw."

    IT was right: the tropical drink still comes out of the straw attached to the virtual desktop. But, you have to wait until the drink gets concocted and mixed in the data center (not on your desktop)...THEN you have first suck through the OS, applications, and user data before the refreshing drink comes through the straw.

    If you don't want your end users to get blue in the face, avoid boot storms altogether by virtualizing desktops using a client hypervisor, and managing the locally-executing VMs centrally. This un-VDI approach is called Intelligent Desktop Virtualization (IDV), and its the best way to avoid boot storms...and enjoy tropical drinks. ;-)
    Dan Cote - Virtual Computer