Accelerating the SAN, with the Server

Posted on September 12, 2012 By Jeff Boles

RssImageAltText

Just a few short months ago, a number of vendors in the market started introducing flash-based, in-the-server, storage acceleration technologies that we label Server-based Storage Accelerators (which I also refer to as “accelerators” in this article).

Across the board, such devices were primarily PCIe form factor cards with on-board flash memory that plugged into a server motherboard slot. Through driver or other software inserted into the server OS’s software stack, the solution would intercept I/Os, cache data from SAN-attached disk onto a NAND-flash cache, and then redirect future I/O requests to that cache.

QLogic takes an approach that may make these accelerators the most disruptive performance technology yet. Why? Because this introduction is likely to make server-based storage accelerators more widely deployable than ever before, bringing the benefits of flash optimized I/O into the SAN.

First, let’s take a look at why these accelerators are a key performance technology to begin with.

Acceleration and the Options

Over the past two decades, performance has been the bane of the enterprise storage administrator, and performance has recently seemed poised to severely constrain the virtualization, consolidation, and data analytics initiatives driving today’s digital business. While data density and compute horsepower have increased almost logarithmically over the years, storage performance improvements have been comparatively miniscule.

In the midst of this clear performance gap, the obvious answer seems to be flash-based storage media. Flash media is highly available in the market, and promises to perform much faster than any rotational disk under the typical small, highly random I/O enterprise workload.

Server-based Storage Acceleration is about inserting flash performance in the I/O path of the server as cache instead of isolated storage in order to achieve maximum performance with lowest latency, and without breaking the shared storage infrastructure. By inserting a flash-based cache, these accelerators do several important things.

First, they offload bursty I/O from shared storage, allowing an expensive storage array to serve up more data and drive up capacity utilization. The best of these technologies cache writes as well as reads, helping to offload even more I/O and decrease latency even further for the application.

Second, these accelerators help shared storage go further by better managing the I/O that gets through to the array. Instead of bursty and random I/O, accelerators aggregate writes and make them steadier and more sequential so that shared storage doesn’t have to work as hard.

Third, accelerators increase application performance without disrupting traditional storage management, because permanent data remains on shared storage. These devices that perform write caching can routinely, or on demand, flush cached data back to shared storage. Snapshots, backups, and other data management can all work as intended.

Since these accelerators come with a relatively low price point, they also deliver tremendous bang for the buck. As we’ve pointed out in recent Taneja Group reports, based on a surface assessment, Server-based Storage Accelerators may push the cost of an infrastructure down to as little as 10% of what would normally be spent to deliver an equivalent number of workloads at the same performance level. This assumes the accelerator cost effectively combines both the I/O processing and flash media, and doesn’t require separate purchase, installation and management of each. Given the total set of benefits, accelerators have rapidly become a key part of the storage performance market.

QLogic – Advancing the Server-based Storage Accelerator

When Taneja Group first wrote about Server-based Storage Accelerators several months ago, no Enterprise Network Adapter-based device yet existed, although we remarked they were surely soon to come. QLogic’s Mt. Rainier technology is just such a solution.

With a combination of enterprise adapter, flash/SSD storage, optimized driver and on-board firmware intelligence, this Server-based Storage Accelerator is an enhanced network HBA that captures all I/O and seamlessly redirects it to flash media that is attached to a PCIe-flash storage card or SAS/SATA attached SSDs via a SAS I/O daughter card.

While Mt. Rainier will be introduced as an FC HBA, it is worth noting that the accelerator could be used to accelerate other protocols, including 10GbE, iSCSI and FCoE on a CNA, but quite possibly also including protocols that don’t even pass through the adapter.

The ENTERPRISE NETWORK AdapteR storage accelerator is more than a sum of the parts. This technology announcement is notable as it opens the door to a broader introduction and acceptance of Server-based Storage Accelerators, and we believe it will act as a catalyst and help resolve several fundamental challenges that stand in the way of dedicated separate accelerators.

Specifically, combined Enterprise Network Adapter accelerators may have several unique capabilities:

Uniquely proven interoperability. Enterprise Network accelerators (HBAs, CNAs) in servers will bring with them all of the historic testing and proven interoperability that HBA vendors are well known for. Such accelerators will be ready to deploy in the infrastructure, with no doubt, and this will lend broad credibility to the technology.

No additional software required. For certain, every Enterprise Network Adapter has a driver, but Enterprise Network Adapter accelerators look likely to use the very same unified driver stack currently used by the adapter (HBAs as well as CNAs). This means that large enterprise customers can look to deploy these accelerators without worrying about software deployment and support of additional drivers and caching SW.

Clustering and high availability prepared. Enterprise Network Adapter vendors also have a long history with clustering and high availability practices in the SAN, and may be quicker to offer broad support for these technologies, many of which can be tricky once caching is inserted in the I/O path.

Ubiquitous availability. The Enterprise Network Adapter accelerator could conceivably become the defacto HBA/CNA in customers everywhere. Adapter vendors may decide that the additional functionality should be present in every product, ready to be licensed and turned on upon demand.

What is your need for Server-based Storage Accelerators?

Server-based Storage Accelerators are here to stay because they solve performance problems without redesigning the infrastructure. If you’re not convinced that Server-based Storage Accelerators have a place in your data center, a few last examples might just convince you. The reality is that Server-based Storage Accelerators have big implications for real world businesses, and stand to fundamentally impact business capabilities like few technologies ever have before.

For example, opening up I/O acceleration for a much broader range of virtual servers and clusters will dramatically increase VM density in the data center. This will come with tremendous payback, as VM density drives overall infrastructure licensing.

The licensing cost alone will justify deploying accelerators in nearly every server. Additionally, managing a single device driver is much easier than multiple drivers and caching SW on each VM.

Another example: with more businesses staring down demands for big-data-like analytics, the performance of many database systems is no longer up to par. Engineering performance for the database has always been tricky and expensive, often requiring expensive dedicated storage or entirely new database solutions.

Server-based Storage Accelerators may democratize storage performance, and thereby democratize big data power. This will make the business use of more data possible within businesses that couldn’t dream of big data power before.

Taneja Group believes flash storage is triggering a fundamental shift in compute power in a more meaningful way than even the Moore’s law defying scaling of the x86 processor over the past decade. But the ideal integration of flash with today’s data center has been on the horizon for the past couple of years, with no clear indication of when the ideal solution might arrive.

In our view, Server-based Storage Accelerators are one of the best solutions yet introduced, and are likely to spearhead a fundamental performance alteration going forward. These broadly available Server-based Storage Accelerators represent an enormous leap into the future of democratized and ubiquitous data storage power. It is about time.


Comment and Contribute
(Maximum characters: 1200). You have
characters left.