Avere’s World Record NFS Performance in the Industry’s Smallest Footprint

I’m blogging from the floor of the SC11 Exhibit Hall where we just broke out the free Red Bull performance drinks and unveiled a new booth backdrop to celebrate our new World Record. Stop by booth 442 if you are at the show.

Avere Systems shocked the storage world today and took the top NFS performance spot on the SPECsfs2008 benchmark, taking down the big dogs, NetApp and EMC/Isilon, in the process. Avere posted throughput of 1,564,404 ops/sec, which is the highest ever posted in the long history of the NFS benchmark. In addition, this throughput was achieved with an ORT (overall response time or latency) of just 0.99 msec, which is 35% better than NetApp’s best and 61% better than EMC’s best.

For more details on the performance tests by the three vendors, here are links to the posted SPECsfs2008 results from Avere, NetApp, and EMC/Isilon.

The performance that Avere demonstrated is impressive but it is only half the story. Even more impressive is the efficiency with which Avere delivered the results.

Avere delivered higher performance and lower latency with a system that costs dramatically less, both in terms of the capital expenses to purchase the system and operating expenses for space, power, cooling, etc.

I will take you through the numbers in a second but first let’s take a look at pictures that compare the storage systems that were tested by Avere, NetApp, and EMC/Isilon.

As you can see, Avere packs the highest performance and lowest latency into a package that is 79% smaller than NetApp and 65% smaller than EMC/Isilon.

Overall size is an approximate measure of the capital and operating expenses. Let’s dig deeper into the actual numbers. In the below table I have included the pertinent comparison data. As you can see from the below, Avere is 51-77% less cost, requires 56-78% fewer disks, and occupies 64-76% fewer rack units.

SPECsfs2008 does not measure power or cooling requirements. In a storage system, the disk drives are the largest consumers of power and dissipaters of heat. Therefore, a good estimate for the power and cooling savings is the disk savings, where Avere is 56-78% less.

Prior to Avere’s SPECsfs2008 posting, NetApp and EMC/Isilon waged a war of words contrasting their SPECsfs2008 results in the body and comments of this blog. It’s a highly recommended read. Make sure to read the comments. With the Avere results now out, the NetApp and EMC/Isilon battle is over 2nd and 3rd place, with Avere taking 1st in all the major categories.

Hope to see you at SC11.

SPEC® and the benchmark name SPECsfs®2008 are registered trademarks of the Standard Performance Evaluation Corporation. Competitive benchmark results stated above reflect results published on www.spec.org as of Nov 15, 2011. Above we compare all SPECsfs2008_nfs.v3 results that achieved 1,000,000 ops/sec throughput or higher. For the latest SPECsfs2008 benchmark results, visit http://www.spec.org/sfs2008.

Jeff Tabor

NAS Optimization for the Cloud

There’s lots of buzz in the storage industry about the cloud. To date, however, the cloud has been impractical for most primary applications because the high-latency WAN connection between the cloud providers and the cloud clients has resulted in poor performance. That’s where Avere comes in…

Avere’s NAS Optimization enables using the cloud for primary applications. The Avere FXT Series uses intelligent tiering to automatically store active data near the client, eliminating the latency of the WAN. Customers are using our cloud solution in four data access scenarios:

1) Remote Office to Datacenter: Data storage is consolidated in a centralized datacenter and Avere is used at the edge to provide low-latency access to remote users.

2) Datacenter to Datacenter: Avere enables compute resources to be shared across multiple datacenters by automatically placing the data actively being processed near the compute nodes and eliminating the WAN latency between the datacenters.

3) Enterprise to Compute Cloud: Enterprises are deploying lower cost compute infrastructures, both in private cloud and public cloud models, by co-locating Avere clusters with the compute nodes to automatically tier and store the active data.

4) Enterprise to Storage Cloud: Avere enables storage clouds for primary applications by placing Avere clusters near the clients, whether in a datacenter, a remote office, or a cloud compute facility, to automatically tier and store the data that the client is actively using.

Check out our cloud solution brief for more.

Enterprise-wide NAS Cloud

Jeff Tabor

NAS Optimization for VMware

VMware is everywhere and I’m very excited about the traction we’re seeing here. Across software build, database, virtual desktop, and other guest applications, customer are finding great value in placing an intelligent read/write caching tier in front of their existing NAS systems. Here are some highlights I’ve heard from customers.

• VMware is a write-heavy app and Avere’s write caching provides a huge performance boost.
• Avere block-level caching efficiently uses SSD/SAS tiers of FXT clusters, especially when using VMware Linked Clones.
• Avere’s NAS Optimization is faster than SAN and simpler to manage and scale.
• The Avere user interface provides great visibility into VMware operations, including ESX hosts, VMs, and VMDKs.
• Storage VMotion makes adding FXT clusters in front of existing NAS systems simple and non-disruptive.
• Storing VMDKs on inexpensive & high-density SATA and accelerating with Avere is much cheaper than storing all VMDKs on SAS/FC.

Check out our VMware solution brief for more.

Avere is VMware Ready certified.

Jeff Tabor

Global Namespace and the Path to NAS 2.0

In the little over a year that we have been shipping our FXT Scale-out NAS Appliance, we have received very positive feedback on our product and its ability to scale NAS performance.  Performance scaling is the result of both our Tiered File System (TFS), which dynamically allocates frequently used blocks to faster storage media, and our clustering technology which clusters up to 25 appliances together to linearly scale performance.

Our primary objective has been to increase NAS efficiency by off-loading filer operations and facilitating the use of high density, low power media for mass bulk storage.  It is common for our customers to perform the same or more processing as traditional NAS with 1/5th of the data center resources (rack space, power & cooling).

Our customers typically characterize our product as the “user or client facing side of NAS” and the traditional filer as the “archive or data management part”.  The most frequent request from our customers has been “Now that you implement the client facing side of NAS, can you do something to help with our NAS clutter?”

NAS Clutter

Storage administrators traditionally have scaled their environment first by adding expensive, power hungry and low density performance disks behind a single filer until that filer becomes over loaded, and second by adding more filers to their environment.  Over time, this results in NAS clutter – a multi-filer environment in which the user or client machines must be aware of (and re-configured with) any changes in the storage environment.
http://averesystems.filhttp://averesystems.files.wordpress.com/2011/03/nas1-0clutter.png?w=400&h=290&h=200
Click here for the full-sized image

NAS clutter is the result of the current NAS 1.0 architecture that was not built to scale performance or handle the challenges of geographically distributed users.  The NAS filer is the single bottleneck in the NAS environment – all operations from all users must transit the filer, much like single CPU processors were the bottleneck in computers until the advent of multi-core architectures.  The NAS 1.0 architecture worked well a decade ago but has severe limitations today.

http://averesystems.filhttp://averesystems.files.wordpress.com/2011/03/nas1-0architecture2.png?w=400&h=290&h=200
Click here for the full-sized image

Towards NAS 2.0

In our product announcement last week, we introduced global namespace, or GNS, functionality to the FXT product line.  Using GNS our customers can now create a single logical namespace in the FXT cluster, which is visible to all clients that mount any FXT Appliance.  The storage administrator can then configure any export on any filer in their data center to be a sub-tree within that namespace.  A single common view is presented to all users or clients, effectively eliminating NAS clutter.
http://averesystems.filhttp://averesystems.files.wordpress.com/2011/03/nas2-0.png?w=400&h=290&h=200
Click here for the full-sized image

Global namespace and the virtualization of storage resources is an important building block for scaling out the NAS architecture.  When you combine global namespace with Avere’s dynamic media tiering and scale-out clustering you have the genesis of NAS 2.0:

  • Global namespace removes NAS clutter from the user view – separating the client facing NAS services from datacenter administration.
  • Dynamic media tiering and scale-out clustering hide mass storage and WAN latency, facilitating the use of high density (low cost, low power) media and remote Cloud storage.

NAS 2.0 and Cloud Storage

NAS 2.0 provides the right combination of global namespace and performance scaling to finally make cloud storage a reality for enterprise applications.  Current cloud storage deployments are typically relegated to backup and data protection applications, due to the high latency to transit the WAN.  Because of that latency, enterprise application performance would suffer and live users would see unacceptably high latency to their data.

Avere’s performance scaling permits enterprise applications and end users to access remote storage with no degradation in performance over local storage.  The deployment model places an Avere FXT cluster near enterprise applications or end users.  Storage can be located anywhere.  The added benefit of GNS in this model is that storage components can be located at several locations with a single access point for all users at all locations – creating a single view of storage for distributed enterprises.  GNS effectively hides the additional clutter of multiple locations for these distributed enterprises.

In summary, GNS is a fundamental component of a NAS 2.0 architecture, whether its within the data center, in the cloud or a hybrid of both. In my next post, I’ll explore more fully how NAS 2.0 enables cloud access.


Storage Tiering is Tops in Storage Magazine’s 2011 Hot Technologies

With less than two weeks left to go until 2011, it’s time for publications to roll out their predictions for the coming year. Storage Magazine has put together a list of what its editors and experts will be the hottest technologies in storage in the coming year. Coming in at #1 is automated storage tiering in its debut on the hot technology annual list, ahead of cloud storage services, primary storage de-dupe and others. Why is tiering so hot? It’s all about the Flash and intelligent use of it:

“It was very difficult to be able to afford enough SSD if you were purely going to use it as a static storage device,” noted Mark Peters, a senior analyst at Milford, Mass.-based Enterprise Strategy Group (ESG). “Now that people will be able to combine tiering with a smaller amount of SSD, I think the two go hand in glove.”

The article goes on to give out sage advice on how to choose a tiering solution and separate hype from reality:

Product differentiators include the level of granularity at which the data moves between tiers, the degree of automation and the extent to which users can define policies.

Wait a minute – that criteria seems very, very familiar to us, it’s just missing the part about the ability to work in a heterogeneous environment.

At any rate, we’re glad that Storage Magazine recognizes how important dynamic storage tiering is in building out high performance, cost effective storage networks – we couldn’t agree more. Here’s to 2011!

Rebecca Thompson

SPECsfs2008 – A Year in Review

11Nov2010 Update: NetApp submitted SPEC’08 results this week on their new FAS6240 and achieved more than 120k ops/sec. They used 288 15k SAS drives to achieve this result so it is just more of the same, throwing disks at performance. They did use 1TB of their Flash Cache but this is read-only cache so they still need a lot of drives for the writes. I updated my table below.

It’s been a little more than a year since my first blog on Avere’s SPEC results and it’s a good time to take a look back to see what’s happened in the past twelve months. In my original blog I used the SPEC results to examine the top-performing NAS systems and compare the number of hard disk drives each requires to deliver the performance.

Why compare the number of hard disks? There are two reasons.

First, the number of disks is a good measure of the cost of a storage system, both the capex acquisition cost and the ongoing operational cost for power, cooling, and rack space. SPEC does not require posting the price of the NAS system under test and the number of disks used is the best way to approximate the price.

Second, there has been a lot of industry buzz around storage tiering, that is, placing data on the best storage media to optimize performance and cost. Comparing the number of disk drives used is a great way to see how each vendor is progressing in this area.

Back on October 12, 2009, the date of my original blog, there were four solutions that achieved roughly 120k ops/sec or better: Avere, Exanet, Huawei Symantec, and NetApp. Since then, six more solutions with results higher than 120k have been posted on the SPEC website. See the below table for the complete list. In the table, I also included the number of disks used by each vendor and calculated the ops/sec per disk used, with this last number being the best measure of performance per dollar delivered by the vendors.

Based on the above results, Avere delivers on average seven times more performance per disk used than the other vendors. Avere FXT appliances use intelligent tiering algorithms to separate performance scaling from capacity scaling and more efficiently deliver both.

Ray Lucchesi, President and Founder of Silverton Consulting, analyzed the SPEC results and reached similar conclusions in his blog last week. For more information, check out his RayOnStorage blog.

SPEC® and the benchmark name SPECsfs®2008 are registered trademarks of the Standard Performance Evaluation Corporation. Competitive benchmark results stated above reflect results published on www.spec.org as of Nov 1, 2010. Above we compare all SPECsfs2008_nfs.v3 results that achieved 120k ops/sec throughput or higher. For the latest SPECsfs2008 benchmark results, visit http://www.spec.org/sfs2008.

Jeff Tabor

Part 4: Things to Consider Before Upgrading Your NAS

If you’re at all concerned with the scalability of your infrastructure when considering upgrades, you should know that adding new filers, more high speed drives and/or Flash modules to a NAS installation to improve performance is a short-term solution at best. It’s only a matter of time before application demands once again outstrip the NAS infrastructure’s ability to scale performance and you’re back to ripping out old gear and replacing it with new. In contrast, with Avere’s two stage NAS architecture, system scalability is built in. As more clients and new applications are added to the mix (requiring higher IOPS performance) an Avere FXT cluster can be easily expanded with the non-intrusive addition of new nodes. Up to 25 appliances can be added to a cluster, delivering plenty of horsepower without having to touch any other devices already in place. And because the Avere FXT cluster can serve multiple storage servers, there is no need to add Flash to each and every filer – the Avere cluster becomes an extensible fast media layer in front of all of them, serving up performance to hot spots without over provisioning.

Manageability is another hidden cost to upgrading an existing NAS infrastructure. With falling prices and improved durability making new storage media such as Flash SSDs widely available to boost application performance, many companies are tempted to install Flash at tier zero and expect that it will solve their performance problems, albeit it at a relatively high cost. But installing fast-access storage media solves only part of the problem. IT then has to figure out which applications are best served by the new tier of storage, often having to become an expert in the latest storage media read and write rates and application QoS schemes in order to optimize the utilization of the more costly storage media. In comparison, an Avere FXT cluster has the intelligence to dynamically allocate data to the appropriate storage tier and media based upon both data and access characteristics, which balances the cost/performance equation with no administrative overhead.

Rebecca Thompson

Part 3: Things to Consider Before Upgrading Your NAS

With traditional NAS, controller upgrades are part of the typical lifecycle of the system. At the time of the initial purchase, the controllers are selected to deliver on the performance requirements at that time but not much more since that would mean spending more money. A year or two down the road, the controllers are out of gas and something must be done to get more performance for the data.

At this point controller upgrades are the typical course of action. Controller upgrades are so common that people have become numb to the pain. Let me remind people of the problems that come with this and propose a better way.

Controller upgrades with a traditional NAS system involve the following steps:
1. Purchase higher performance controllers (typically a model or two up the traditional NAS vendor’s product line, if a higher model exists)
2. Purchase new disk shelves or Flash-based PCI cards, required to get more performance out of the new controllers
3. Purchase new licenses for all software (e.g. NFS, CIFS, mirroring) at the higher price tier of the new controllers
4. Take the original NAS system offline
5. Remove the old controllers
6. Install the new controllers
7. Add the new disk shelves
8. Bring the upgraded system back online

The above process is expensive, disruptive, and requires more disks, power, and space even if no additional storage capacity is required. As an alternative, consider boosting the performance of your existing NAS with Avere FXT appliances.

Avere FXT appliances boost the performance of all NAS applications by accelerating read, write, and metadata operations without the need to add more disks to your NAS system. FXT appliances scale from 1 to 25 nodes per cluster to match your initial needs and gracefully scale as your needs grow. If more capacity is needed, then this can be accomplished by adding cost-effective SATA drives to your existing NAS system. FXT appliances are simple to install in existing environments and require no changes to existing applications, clients, NAS systems, or data retention procedure such as backup and mirroring. FXT appliances accelerate NAS performance and enable a 5:1 reduction in disks, power, and space when compared to traditional NAS.

With Avere you get the benefits of a clustered storage solution today without the need to migrate your data to a new storage system. For more information on FXT appliances, please visit the Avere website product page.

Jeff Tabor

Part 2: Things to Consider Before Upgrading Your NAS

Integrating Flash storage into NAS systems is a new method of improving performance that is being promoted by vendors of traditional NAS. List pricing for Flash in a NAS system from one of the leading vendors runs from $170/GB to $300/GB. This compares with roughly $2/GB (again, list pricing) for SATA storage. At these prices, customers need to be using Flash very efficiency. Sadly, vendors are not always making this possible.

Typically, Flash is added to a NAS system in one of three ways, as a PCI card, an SSD array, or a caching appliance. Let’s look at the leading vendor in each category and the efficiency of their approach.

NetApp offers Flash as a PCI card they call PAM. Up to five PAM cards with a total of 2.5TB of Flash can be installed in some NetApp controllers and pricing is in the $170-200/GB range. PAM is inefficient for two reasons. First, PAM is read-only so customers still need many hard drives to handle the write workload. Second, PAM can only help the read performance of the controller in which it is installed. This means you may be adding up to five PAM cards to every NetApp controller in your environment. This approach is expensive and highly inefficient in environments with multiple NetApp systems since Flash is added to individual storage “silos” to handle the peak load on the silo but is under utilized much of the time.

EMC offers Flash as an SSD array. Up 16 SSDs and 6.4TB of Flash are supported per array and pricing is in the $200-300/GB range, making a fully populated array very expensive. Beyond the price, EMC’s SSD arrays are inefficient since data movement to Flash is slow and not granular. Data movement is triggered by a policy engine that measures data activity across long periods (e.g. days) and cannot immediately respond to a hot application. Also, data is moved in large volume-level chunks meaning lots of cold data is using expensive Flash storage alongside the hot data.

Avere offers Flash as a caching appliance that sits in front of NAS systems from other vendors. The Avere FXT 2700 has 512GB of Flash per appliance, scales to 25 appliances (13TB) per cluster, and accelerates read and write workloads. The Avere architecture is highly efficient since it provides a consolidated Flash layer that is shared by all the NAS systems. This allows customers to provision the right amount of Flash to deliver the aggregate performance across all the NAS systems. In addition, the FXT 2700 makes the most efficient use of the Flash storage by moving data in real-time and at the finest level of granularity possible, blocks with files.

For more information about the FXT 2700 and other FXT models, please visit the Avere website product page.

Jeff Tabor

Part 1: Things to Consider Before Upgrading Your NAS

In today’s homogeneous-media NAS architectures, users will invariably be asked to add hard disk drives to increase storage capacity or system performance. Beyond the acquisition cost of the hard disk media, users need to consider the total cost of adding hard disk drives to a NAS infrastructure. Beyond the acquisition cost of enterprise-class Fibre Channel or SAS drives, you also have to factor in the hidden costs of power and cooling, the necessary rack space and data center floor space. Furthermore, adding hard disk drives is a costly and largely ineffective method of increasing NAS performance. HDDs are inefficient at providing IOPS, so adding expensive HDDs is not a good IT investment from a cost-performance metric.

The value proposition becomes even worse when companies over provision capacity and short-stroke drives in a futile attempt to achieve higher performance. Instead of trying to extend HDD technology, NAS users will get a better return on their storage investment by moving from a homogeneous solution to a tiered NAS solution. Adding intelligent storage tiering, that makes the best use of newer solid state disks, produces a dramatic gain in IOPS performance, with far lower impact on power, cooling, rack space and data center floor space. This approach eliminates the expense of adding potentially hundreds of enterprise HDDs to the NAS environment, recovers the capacity on existing HDDs that was previously unavailable due to short-stroking, and solves NAS performance issues while enabling IT to leverage low-speed (and low cost) SATA drives for capacity requirements.

The two-stage tiered NAS architecture, using Avere’s FXT family of products, provides for all of these benefits by permitting the addition of solid state media to any existing NAS infrastructure.

Ron Bianchini

Follow

Get every new post delivered to your Inbox.