Danish food supplies company Chr. Hansen deployed PernixData FVP flash caching software for around €19,000, avoiding a spend of up to €250,000 in extra software licence fees or the need for a very costly SAN fabric upgrade.
Chr. Hansen supplies cultures, dairy enzymes, probiotics and natural colours to the food, health and animal feeds industries.
It has three datacentres that support seven production sites and 30 sales offices in 37 countries. Key applications are Microsoft SQL, SharePoint and Biztalk plus Oracle and SAP, along with homegrown systems for factory control and research and development (R&D).
Most of the Chr. Hansen IT environment is virtualised on VMware, with around 1,900 virtual machines. Storage is a mixture of IBM StorWize V3700, V5000 and V7000 arrays, with around 250TB of data retained.
Read more about flash caching
Mid-way through 2013, Per Helsbøl, IT infrastructure specialist and architect, started to see symptoms of the main SQL and Oracle database cluster suffering performance problems. Applications that used them “felt sluggish” and there were issues with data processing.
He said: “Average response times were 4-5ms, or 15ms at peak, but when latency hit a norm of 15-20ms we started to see a lot of application users complain.”
Upon investigation, the bottleneck proved to be the host bus adapters (HBAs) connecting a cluster of two servers (running 32 cores) and their back-end storage arrays over the SAN fabric. (An HBA is a host networking card enabling connectivity to the SAN).
The options available to deal with the issue were to double the size of the server cluster – and also the number of HBAs – to add 32 more cores in two more hosts, or to upgrade the SAN fabric from 8Gbps Fibre Channel to 16Gbps.
“Moving to 16Gbps Fibre Channel would have helped but it would have been a huge cost, especially when our last Fibre Channel upgrade is not yet written off,” said Helsbøl.
To extend our SQL cluster we were going to have to invest approximately €250,000. By investing a far smaller amount in PernixData FVP software and some additional, high-performance SSDs, the return on investment case was obvious
Per Helsbøl, Chr. Hansen
Meanwhile, the addition of 32 cores in two new server blades was set to cost an extra €250,000 in VMware, Microsoft SQL and Biztalk licenses.
At around this time Helsbøl visited VMworld in San Francisco and came across PernixData, which provides flash caching software.
Helsbøl said: “This meeting resulted in running PernixData FVP software in a test environment. The trial demonstrated good performance, even on slow SSDs.”
“To extend our SQL cluster we were going to have to invest approximately €250,000. By investing a far smaller amount in PernixData FVP software and some additional, high-performance SSDs, the return on investment case was obvious.”
Instead of a projected €250,000 spend, Chr. Hansen’s PernixData deployment cost around €9,000 for flash drives and €10,000 for PernixData licences. These were deployed on a new server cluster, the cost of which was accounted for by existing upgrade plans.
PernixData FVP software has delivered improved application performance results, said Helsbøl.
He said: “We now see faster response time on all SQL and Oracle servers, with less I/O on back-end storage. Production reports that took between one and two hours to run are now down to 15 to 25 minutes and our average response time on SQL and Oracle servers went down from 12ms to less than 2ms.”
PernixData FVP is deployed in the VMware hypervisor, from where it intercepts database writes and sends them to local flash storage. The data can then be used by the database from that flash cache, while the data is also written to the storage array.
PernixData is one of a number of caching software products that aim to speed access to data.
PernixData’s product creates a form of write-back cache, where data is acknowledged to the host before it is written to permanent storage. This can risk data loss in cases where a server fails but, with PernixData FVP, customers can specify that data is written to more than one cache to mitigate against host failure.