Server Side Caching: A VMware Breakthrough

Posted: 01/20/2012

I have met with many CIOs over the last year and have found that they are managing several somewhat conflicting objectives:  improve performance, decrease costs, manage more equipment with less people, consolidate data centers, consolidate servers, improve customer service, and preserve business continuity in the event of equipment or site failures. 

One of the main tools that CIOs have deployed to help them reach these objectives is server virtualization.  Server virtualization deployment, typically with VMware, started in trials with non-critical environments and progressed to supporting most back-office IT applications (Exchange, Sharepoint, etc).  With the back-office virtualized, most organizations are looking at ways to extend virtualization to support mission critical applications such as those running on Oracle and Microsoft SQL Server.  In this way, VMware is the giant black hole in the center of the IT universe, which having consumed the centrally located back-office IT environment is generating the gravitational pull to grab hold of those line-of-business applications that have historically run in physical server environments.

CIOs are plenty savvy.  Performance and scalability demands from line-of-business executives prompt them to keep these apps on separate physical server environments, but known capital and operational cost reductions from virtualization push them towards centralizing.  Can they get the best of both worlds? They know VMware is hard on database performance, but they can’t help wanting to virtualize  the database servers as well because it just makes their life easier and offers significant savings on the server infrastructure.

Once a database environment is virtualized and sharing precious server, network and storage resources with other applications it is increasingly hard to improve application performance.  The market is hungry for a way to make mission critical VMware applications faster, but the enterprise shared storage arrays incorporating flash are too expensive and generally involve fork lift upgrades of their existing environment.  Given the current economic climate, fork lift upgrades are unlikely.

So what can a CIO do to improve VMware performance without forklift upgrades?  One of the more intriguing new methods for improving application performance in virtualized environments is to use server caching with flash memory.  In 2011 alone, Fusion-io acquired IO Turbine (an innovative developer of server side caching software), STEC introduced caching, and EMC announced Project Lightening.  These products are likely just the beginning of a long string of products in this category.  However big you view the problems related to VMware performance today, I am convinced it only gets bigger as more applications get moved into virtualized environments.

The reason server side caching is such a breakthrough in VMware is that it is seamlessly integrated, it is lower cost than other Flash memory solutions, it actually makes shared storage faster by offloading reads, and it does not disrupt array-based replication. 

Server side caching starts with installation of a PCIe Flash card in the VMware server.  For this to work, the software has to be VMware integrated and has to be aware of the guest operating systems.  At this point, the server side caching software allocates flash memory to each guest operating system as a read only caching layer.  The read only caching layer is “warmed” by monitoring reads and writes out of the guest OS.  Over time, many reads are served from the caching layer.  The colocation of the PCIe flash card with the VMware guest operating systems means that cached reads are served up at the lowest possible latency giving an immediate boost to application performance.  Reads that come from server side caching never hit shared storage which has a side effect of decreasing the load on shared storage meaning that the storage you have is more efficient.  Low cost is achieved by pairing the caching software with MLC-based flash memory.  This is far more cost efficient and productive than buying SLC-based flash and placing it in a shared storage array.  Finally, as the server side caching layer is only caching, reads and writes still make their way without interruption to the shared storage for array-based replication.

Server-side caching is an important new tool for CIOs and data center architects looking to cost-effectively improve performance.