Last month I wrote a post about the new virtualized Sharepoint solution using Microsoft Hyper-V on EMC VNX. That particular solution focused on the tiered storage of the VNX and how to use the storage array for both primary production data to storage SQL databases containing active Sharepoint data and how to use less expensive drives as a target for BLOB data that was no longer active.
Another solution EMC recently developed focuses on testing performance with SQL Server on the VNX. Two primary tests were performed, one centered on testing the performance of a SQL Server database in a physical server configuration against the VNX and then comparing this to a virtualized configuration using Microsoft Hyper-V. The second test measured the number of Transactions per Second (TPS) for a SQL Server database against the VNX and then compared the results when FAST Cache was enabled on the array. For this second test the databases were only tested in a physical server configuration.
Let's first look at the environment. The following diagram shows the test configuration used:
This solution contains a single Microsoft SQL server running against the EMC VNX5700 array. Both the transaction log files and the database files are using LUNs from a RAID 10 group. The SQL Server database is a single instance, 400GB single database with an OLTP transactional based workload. The SQL Server is connected to the array via a Fibre Channel connection.
The VNX array is using RAID 10 300GB SAS drives (15k rpm) for the SQL Server databases. There are also 4 100GB EFDs (Enterprise Flash Drives) in a RAID 1 configuration for the FAST Cache configuration. FAST Cache expands the read and write cache available on the storage array by using EMC EFD devices as a cache location. Testing like this one has shown random data access is much faster with EFDs then accessing data using similar Fibre Channel drives.
So, let's get to the results. What did we see? For the first test in which we tested physical vs virtual on the VNX (including both Pass-through for Hyper-V as well as VHDs) we saw that the VNX performed very well in number of Transactions per Second in both a physical as well as a virtual when using Pass-through. While Pass-through resulted in a 5.5% drop in performance compared to physical we still achieved over 1360 TPS. When using VHDs we saw a noticeable drop in performance (24%) compared to physical and Pass-through.
Next we conducted the test to see what advantages we would get in this environment when using FAST Cache. Keep in mind that we already achieved 1448 tps on the physical server environment without FAST Cache. So what happened with the same exact workload when we enabled FAST Cache? 300% performance increase to over 5800 Transactions per second! Yes, you read that right. 300%!! Here is the graph and the supporting data which I'll let speak for itself:
So what was the point of this report? It validated that FAST Cache does indeed have significant performance gains in random I\O workloads. It also showed that Hyper-V Pass-through can perform almost as well as a physical server configuration in this same workload. The document has more information including all of the hardware used for the test, how the environment was configured as well as several other details so take a look at the report and expect some more updates on our virtualized solutions with Hyper-V and VNX!
Great post! Thanks for sharing. I've been wondering how FastCache would help one of my SQL workloads.
Posted by: BrandonJRiley | 04/08/2011 at 08:05 PM