Network and Storage Protocols

How to Achieve Space Efficiency AND Performance with Dedupe AND PAM


Cross post from the NetApp Virtualization Effect Blog:

Subscribe to the NetApp Virtualization Effect Blog for all the latest & greatest of NetApp & Virtualization Technologies.

Posted by Chris Gebhardt - Reference Architect

There has been much talk in the blogosphere lately about NetApp technologies. In particular NetApp’s space efficiency and performance capabilities.  Well I figured that it was time to post some information that may help answer a couple questions that people may have.

We have been conducting research in our labs on how customers can use NetApp space efficiency technologies while providing exceptional performance. Well the problem may sound complex but the solution is actually simple. It's in there!!! The secret sauce is in Data ONTAP & Intelligent Caching! Here are just some of the results that we have captured.

The setup
We used one NetApp FAS3170 head w/ 23 data disks and a 16GB PAM card and booted 1000 XP virtual machines.  We used 8 - 16 CPU servers with 128 GB of memory each running VMware ESX 3.5. We provisioned this infrastructure using NetApp RCU and created 6 - 40GB NFS datastores each with approximately 167 VMs’ each. That’s right, we had 167 VMs’ of 10GB each in a 40GB datastore (with room to spare). Since RCU was used to provision these VMs they were already deduplicated!

We then booted the virtual machines all at the same time with Intelligent caching disabled.  We then used performance advisor to graph the disk throughput, network throughput and NFS protocol latency to see what effects dedupe has on this massive read storm. Here comes the good stuff, we then waited a bit to get half way through the boot storm and then enabled Intelligent Caching with PAM. Here is a screenshot of the entire boot of 1000 virtual machines. As you can see it takes about 12 minutes to boot these 1000 virtual machines. This screenshot is is of the NFS protocol latency and as you will notice the average latency is ~75ms without intelligent caching and ~7ms with intelligent caching (Thanks @Tintop). At about 15:39 I turn on Intelligent caching and latency drops by 10X!

VMware FusionScreenSnapz003

This next graph shows the exact same time period but for Disk Throughput.  By using NetApp Intelligent Caching in combination with disk deduplication you can reduce the IO coming from disk because not only is the data deduplicated on disk but it is also deduplicated in CACHE!

VMware FusionScreenSnapz004

And just as another data point here is what the network looks like when running this boot test as well...

VMware FusionScreenSnapz005

Using NetApp deduplication can not only reduce the storage footprint but allow customers to serve data more efficiently with NetApp Dedupe Aware Intelligent Cache.  This allows customers to reduce the number of spindles needed to everyday workloads and handle peak workloads such as boot storms while delivering a robust end user experience. you would like to read more on NetApp Deduplication and Performance please take a look at Vaughn's blog over at The Virtual Storage Guy

Follow me on Twitter@chrisgeb



Beautiful -- fantastic graphs and in its own way more useful than a TR even. (short and sweet to send on as a link to people curious about PAM benefits/impact)