Is Your Tiered Storage Like an Idling Ferrari? By Doug Rainbolt On August 22, 2011
In a previous posting I addressed NAS “hot spots” and closed by stating that I would follow-up with a post related to tiered storage. The concept of tiered storage is relatively easy to understand. All data is not the same, whether we’re talking about its intrinsic characteristics or the frequency of access requests to it. Having data reside on different media based upon performance requirements makes sense. In general, data that’s hardly accessed should be on different media than data that is highly active. So what is there not to like? I’d like to share a couple of thoughts: First, it’s important to understand your goals when considering tiered storage. Second, thinking of a Ferrari as analogous to Flash or Tier0, it’s important to understand if your “city streets”, like the hosting NAS controller or node, even allow for real acceleration.
Let’s start with your goals. All data is not created equal. Obviously this is a platitude. But I would argue that storage array designers, although acknowledging that this is true, over designed systems such that any data, hot or cold, could benefit from performance. An extreme example of this was the use of massive numbers of disk drives that are underutilized in the attempt to increase IOPS. Not all of the data stored requires this kind of performance, but it was there just in case. Not the most efficient model. Fast forward, tiered storage “resurfaces” as a storage model, making use of Flash as a tier or cache for active data and lower cost media such as SATA drives for less active data. If your goal is to achieve better efficiencies, this model certainly helps. However, if your goal instead is to achieve both improved efficiency and effectiveness as measured in incremental improved performance that can benefit your bottom-line, you need to take a closer look. This takes us to the Ferrari.
When you look deep in the architecture of a storage array, it’s like looking down on New York City at night. Cars, bikes, and busses are moving everywhere. And a Ferrari at a red stoplight in this city is going nowhere. There are traffic conditions in server architectures that can stymie traffic flow, no matter how fast your wheels are. Within NAS servers there are multiple processes interworking. Let’s assume that data has moved between media tiers. An NFS request comes in, the NFS daemon discovers that the data is no longer in memory; it’s now sitting in Flash. A DMA is executed and while this happens, the NFS daemon goes to sleep. The entire stack for this transaction is saved. After the data is moved, the daemon wakes and the request is completed, starting at the point where the sleep process was invoked. The more dynamic the data is, moving from tier to tier, the more the controller must do to switch between processes. This adds latency at the system level. The trick is to get data in and out of memory very quickly while consuming minimal controller processor and networking bandwidth. The high IOPS potential inherent in Flash is inhibited if data isn’t quickly put on or received off the wire, much like the Ferrari. What may have looked like a free lunch isn’t so free. Don’t focus solely on the media tiers the vendor is embedding, you should also understand at the system level how the product will scale as your data grows.
If your system can’t cost effectively scale, delivering compelling cost per I/O at the systems level, you’re going to be adding more and more resources. Don’t settle for believing that tiered storage is simply about embedding different media and executing automated data movement. It also includes a processor, networking, and memory elements that enable data movement.
The idea behind a performance layer appliance, such as the ANX 1500, is to accelerate data movement between NAS, the ANX 1500’s memory, and Flash with minimal system resource. This is because the ANX 1500 features a single stack design and makes use of data acceleration technology such as TCP Offload that avoids the need for sleeps, wakes, or interrupts that are so characteristic of traditional server design. The potential of Flash is unleashed to both make efficient use of media and to effectively increase performance. This translates into completing your work faster and getting paid sooner. An NFS acceleration appliance actually enables Flash and other elements of your storage infrastructure to run at their full potential without encountering orange and red traffic lights. And this is done at a fraction of the cost of adding additional tiered storage infrastructure. You can add to the benefits of tiered storage when using the ANX 1500 with existing NAS infrastructure. With the ANX 1500 it’s like your Ferrari is unleashed on the Autobahn rather than traveling down city streets.
In my next posting, I’ll expand what is meant by a single stack design and why it’s no value to I/O acceleration.
Related posts:
Leave a Reply