Join Newsletter
Forgot
password?
Register
Trusted Business Advisors, Expert Technology Analysts

Taneja Blog

Taneja Blog / Data Center Systems

QLogic FabricCache Solves Caching Issues in Virtualized and Clustered Environments

We are excited about QLogic’s new FabricCache offering. As the first product based on QLogic’s Mt. Rainier technology, the FabricCache QLE10000 is a converged Fibre Channel and caching host bus adapter (HBA), which is transparent to hypervisors, operating systems and applications. FabricCache is the industry’s first adapter offering that enables the cache from individual servers to be pooled and shared across multiple physical servers. It’s designed especially to accelerate enterprise apps in multi-server virtualized and clustered environments that rely on SAN storage.

Taneja Group’s just-completed survey of nearly 280 IT managers and executives – including a healthy mix of users and buyers – supports the market need for storage acceleration in general, and server-side storage acceleration solutions in particular. More than three-quarters of the respondents said that they are thinking about deploying a storage acceleration solution of some kind. More specifically, 56% of respondents indicated that server-based acceleration is one of the acceleration approaches they most value, with users overall valuing it more highly than array or network based approaches. But in contrast, only 26% said they had deployed a server-based acceleration solution to date, citing a number of obstacles that have prevented them from moving forward.

We believe that FabricCache adapters will help customers to overcome many of the limitations of today’s server-based caching solutions. Enterprise users have traditionally overprovisioned cache in single servers to ensure that critical applications can achieve high performance levels, even during periods of peak IO demand. This has been especially true in highly virtualized and consolidated infrastructures, in which multiple virtual servers or virtual desktops are contending for the same set of physical SAN resources. The QLE10000 eliminates the need for costly overprovisioning by allowing single-server flash cache to be transparently and coherently shared across multiple physical servers, irrespective of the operating systems, hypervisors and applications they’re running. In effect, this enables FabricCache to serve as a caching SAN adapter, the first of its kind in the industry.

FabricCache is also much simpler for IT administrators to deploy and manage than many of today’s server-side acceleration offerings. With FabricCache, customers do not need to install multiple drivers or a caching agent per OS; all that’s required is a single standard adapter driver, based on the QLogic driver stack that has been proven in use with millions of installed Fibre Channel ports. This is a particularly big deal in virtualized environments, in which virtual server acceleration solutions typically require a driver to be installed in each guest OS. Since the FabricCache adapter appears logically to the server as a single QLogic HBA, it can be integrated transparently and non-disruptively into existing enterprise environments.

One use case in which FabricCache stands out is in the live virtual machine (VM) migration between physical servers, which we saw QLogic recently demonstrate. In this demonstration, FabricCache adapters were used to cache SAN storage during a vMotion-enabled live migration (this migration took place in a VMware vSphere environment, but FabricCache supports other major hypervisors as well). Normally as the result of a vMotion operation, you would expect to see at least a temporary degradation in IO performance during and immediately after the migration, since the migrated VM needs time to re-warm its cache on the new server following the vMotion. But in this instance, FabricCache enabled the shared cache to stay warm, and performance remained high during the migration exercise.  And notably, advanced vSphere capabilities such as DRS workload balancing were not compromised. When you extrapolate the performance advantages from the two vMotions we saw in this demo to a large number of such migrations, you can see the potential productivity and quality of service benefits that FabricCache can deliver to virtual infrastructure users.

FabricCache also provides economic benefits: because it can be integrated non-disruptively into any storage and server infrastructure, running any combination of hypervisors, operating systems and applications, FabricCache eliminates the costs of additional drivers and caching software, and helps to extend the life of an existing SAN. We look forward to seeing how customers apply FabricCache in practice, as it begins shipping in the next few weeks.

Bookmark and Share
  • Premiered: 03/21/13
  • Author: Jeff Byrne
Topic(s): Qlogic FabricCache Storage Acceleration SAN Server Virtualization

Comments

Great blog, Jeff.  Clustered and virtualized environments are very tough to accelerate.  What makes them inherently difficult to cache and why are there so few solutions to address these applications?

By ctbrett on 04/01/13

Caching in virtualized and clustered environments is challenging for several reasons. For one, the IO blender effect of virtual machines consolidated on single systems creates patterns of random IO, which take the predictability out of the IO stream and make many caching strategies less effective.

Secondly, these tend to be very dynamic environments. What happens to a virtual machine cache, for example, when a VM is moved from one physical server to another? Does the cache somehow follow? Some array-based caching approaches allow this, but they are not as efficient as server-side caching solutions, which are closer to the VMs and avoid the need to traverse the network to fulfill cache read/write requests.

But server-side caching solutions have traditionally been restricted to a single server, which prevents them from addressing the performance needs of a typical virtualized or clustered environment. Server-side caching approaches are often complex as well, requiring installation of multiple drivers or agents per OS to implement.

The QLogic FabricCache QLE10000 shared cache solution addresses all of these issues, providing a promising approach to solving what has traditionally been a difficult technical problem.

By Jeff Byrne on 04/15/13

 

Leave a Comment

You must be logged in to comment. Click here to log in or register if you don't have an account.