Our Microsoft Practice Head at DSP talks to us about upgrading one of our clients SQL Server 2008r2 environments to SQL Server 2012 using dsp SQL Server Modernisation & Consolidation strategy. While the results are detailed for SQL Servers, dsphave also had similar results with Oracle workloads including 12c, 11g and 10i instances...
The results of the migration in terms of business as well as technical delivery are pretty outstanding.
As part of the migration, DSP “Stress” the old and new environments to see how fast they can run SQL Server. The best way to measure a server is to count the number of “Batches per Second”. This is essentially the speedometer of a SQL Server and the best way to measure how fast we can make the infrastructure work for SQL.
The server we were migrating from in this scenario was a DL380 G7, which was dedicated for SQL Server with 24 CPU’s 32GB RAM on a shared 3par Storage tier. This server was running SQL Server Enterprise and so cost our client £96,000 in SQL Server licenses.
The solution we architected for our client comprised of DL380 G8 servers which use a Violin 3200 SAN Storage layer. This server was using a hypervisor and had 8 vCPU’s and 32GB of RAM. Now you would expect the new server with 8 cores to be pretty underpowered compared to the server with 24 cores, even though it was a generation behind, and on paper yes, the new server has less compute than the old server.
However, the new server has a better CPU and the Violin storage disk subsystem which makes a remarkable difference as shown in the chart below (with Laptop added as a benchmark).
The New Server (with 8 cores) when put under the DSP stress tests ran 81,000 batches per second while the older server (with 24 cores) only managed 31,000. That is not only a 2.6x improvement in SQL Server performance, but the costs to license this new server is 67% LESS than the older server at £32,000. That’s a 2.6x improvement in speed with 67% reduction in licenses.
Why can this be so much faster? The reality is, SQL (like any application) will only run as fast as its slowest bottleneck – in this case is storage. On the old server with 24 cores, even though we tweaked and optimised the environment, under the stress test we could never get the CPU utilisation above 60%. This is because the storage subsystem was maxed out and throttled the CPU’s back to only 60% of their capacity. That’s fine until you realise 40% of the £96,000 license investment (£38,400!) would in essence be wasted and never used.
With the Violin storage, this is a different story. The storage is built on SLC flash memory and has a microsecond response time (not millisecond) and 150,000 I/O’s. This means the storage could more than keep up with SQL demands during the stress test. This allowed all of the CPU’s to run at 97% in the stress test, resulting in the 81,000 batches per second throughput.
When you look at the investment made into the hardware, the existing Server had over £330,000 of hardware and storage behind it. The Violin Server had around half of that investment. Overall the solution would save around £264,000 in storage and licenses more than justifying the costs to upgrade and migrate to SQL Server 2012.
Solution Profile – SQL Server 2012 Private Cloud
SQL: SQL Server 2012
Operating System: Windows 2008r2
Hypervisor: VMWare 5.0 / Hyper V
Server Hardware: HP DL380 G8 Servers with 2 x 2.13Mhz (Intel Xeon E7-4830) processor and 256GB RAM. The Server has a Storage SSD card for ESX plugged into a PICe slot.
Storage: The storage for the SQL cloud is provided by a Violin 3200 Storage Array. The Violin array is SAN based Memory array providing 4TB of space and 200,000 sustained I/Op a second and a 200 Microsecond (not milliseconds) response time.
To find out more about Violin Storage arrays see here: https://www.violin-memory.com/i/