The following is a post by David Wartell of R1Soft, who will be presenting at HostingCon 2010 on Tuesday, July 20th at 8:00AM.
Virtualization is obviously great business for hosting and cloud service providers enabling new capabilities in measuring service levels, provisioning, and high availability. The big surprise has been that the more we use virtualization, the more we have storage performance issues.
Virtualization is all about sharing and making the most of hardware resources. Let’s face it, we would all like each application to have its own dedicated physical server, but when it comes down to it it’s just a matter of cost and manageability, and all we are trying to do is squeeze as much out of our hardware and floor space as possible.
This all sounds feasible with quad-cores becoming commonplace, memory perhaps getting exponentially less expensive, and 1 Terabyte SATA drives being found in the bargain bin. The big challenge is that the storage has not really gotten all that much faster in the last two decades.
So now take the same disk performance you had with a physical server running a decade ago and share it with five, ten or even fifty virtual machines. You have now been introduced to what people are calling the Virtual Machine I/O Blender. When you had a single O/S to disk ratio, storage performance was relatively smooth, but as you add additional virtual machines the disks have to manage all of these different I/O loads and this causes serious performance problems.
If you make a list of the top things that can be modified to improve virtualization storage performance you have to look at backups as probably number one without a close second. Typically the largest user and biggest waste of disk I/O is the backup process, and the reason is simple. Most backup software reads all or most of the data off of the disks every time it runs.
A decade ago if we were talking about backup software we would be talking about how to save network bandwidth, how to save CPU cycles, and how to get the backup off of the network and server. This was accomplished one of two ways. First, SAN backup where you read the data directly off of disk without involving the server CPU or network, and second the rsync algorithm, an ingenious way to determine small parts of files that changed and only use network bandwidth for the changed parts or deltas. Both of these methods are widely used by most commercial and free backup software backups. Both of these methods do nothing to help your storage performance problems.
The way to check-off backups from the list of virtualization storage problems is to use asynchronous-replication type backup software. What asynchronous replication does is drastically reduce the I/O requests involved in a backup operation. And by “drastically”, I mean it takes 15 minutes instead of 15 hours. This works by using special device drivers to know ahead of time which parts of your file system or virtual disks have changed since your last scheduled backup operation. By doing this, only the data that actually changed is read from the disk, and the rest is ignored because, on average, only a fraction of a virtual machine’s data changes on an hourly or even daily basis. This means very little disk I/O is done during the backup operation. You can learn more about asynchronous replication products at http://www.r1soft.com
Join David Wartell from R1Soft on Tuesday, July 20th at 8:00AM for a HostingCon session where he will discuss how asynchronous backup is revolutionizing disaster recovery and data storage in the cloud.
About David Wartell
David Wartell is the founder of R1Soft. Prior to BBS Technologies, Wartell was the President and CEO of Righteous Software Inc. Wartell has over 10 years experience in software development and Internet infrastructure and has held positions ranging from software engineer and network engineer to CEO. Before starting R1Soft he worked as the Sr. Network Engineer for EV1Servers/ThePlanet in Houston, Texas where he was responsible for one of the largest computer networks that supported over 20,000 Linux and Windows servers.
Attend this session and save before it’s too late! Early bird savings end 5/31/2010!
Register for HostingCon 2010 Now and Save $140 on your full conference registration.
