Scale-out NAS is becoming popular, with most major vendors offering these types of products. As a reminder, scale-out NAS systems will increase performance and capacity at the same time – although you don’t have to scale the systems in the same ratio. You can add controllers for performance, storage for capacity, or both.
There are several things to look at when considering a scale-out NAS system.
1. Is a single namespace provided across all the nodes (also called controllers or heads) so that a file system can be spread across the nodes but the user does not need to take any special action for accessing a file? There are different ways that a single namespace can be implemented, and some may be better than others. Mounting or sharing a file system on a scale-out NAS system should require no more effort than if it was on a single-node system.
2. Does the management software manage across all nodes as an aggregate but still allow individual node communication to detect problems in the system?
3. Is there load balancing across nodes? The load balancing can be automatic when files are stored to distribute data across the different nodes. Will data automatically be redistributed across nodes (in the background) for capacity or load balancing?
4. Can it scale independently? In other words, can you scale nodes for more performance and the underlying storage for more capacity? This provides the greatest flexibility in usage. If the answer is yes, then how many nodes can the system scale to include? And how much capacity (including storage controllers) can it scale to?
5. Is there a back channel for communication between nodes? This requires another communication path between nodes rather than using the same path clients may be using to access data. Examples of this may be an InfiniBand connection between nodes or a 10-Gigabit Ethernet connection. Usually there would be a pair of back channels for availability.
6. Are there any features that are not included that would normally be part of standard NAS systems? A few to consider are: snapshots, remote replication, NDMP support, NFS and native CIFS support, security controls such as Active Directory, LDAP and file locking for shared access between CIFS and NFS, anti-virus software support and quotas.
7. Does the scale-out NAS support both small and large files? Some of the distributed file systems used for scale-out NAS come from the high-performance computing area where the optimization was around large files. It is important to understand whether the system supports small files and large files equally.
This list is a first level look with more detailed differences to be explained in an upcoming Evaluator Group article.
(Randy Kerns is Senior Strategist at Evaluator Group, an IT analyst firm).
Oracle CEO Larry Ellison today answered the question of what he will do with Pillar Data Systems, which he has invested hundreds of millions of dollars of his own money into. Oracle said it has agreed to acquire Pillar and will use its storage as its main SAN platform.
The Pillar deal is likely to be a big topic Thursday when Oracle executive president Mark Hurd and vice president of systems John Fowler host an Oracle storage strategy update that will be webcast.
In a letter to Oracle customers, Fowler referred to Pillar as a leading provider of SAN block I/O storage systems and highlighted its quality of service and scalable architecture. Fowler wrote that Pillar has nearly 600 customers running 1,500 systems, and boasted that the utilization rate of Pillar Axiom systems is about twice the industry average.
A presentation about the deal on Oracle’s website said Pillar Axiom will become one of four Oracle storage products dedicated to running Oracle software better. The others are Exadata Storage Servers for databases, ZFS Storage Appliances for NAS and StorageTek’s tape family.
Most of Oracle’s storage platform was acquired in the Sun deal, and Sun resold SAN storage from Hitachi Data Systems and LSI. After Oracle acquired Sun last year, Ellison said his company would concentrate on selling storage developed in-house rather than OEM products. It dropped its partnership with HDS for high-end SAN systems last year. It continued to sell midrange SAN systems from LSI, but during Oracle’s quarterly earnings call last week CFO Safra Catz said sales of the LSI systems dropped in the wake of NetApp acquiring LSI’s Engenio storage business. Pillar Axiom will likely replace NetApp Engenio systems as Oracle’s main SAN platform.
People in the storage industry have wondered about Pillar’s fate ever since Oracle bought Sun. Ellison’s venture firm sunk $150 million into Pillar to get it started in 2001 and put a lot more in to keep it running over the last 10 years. Pillar didn’t get products out the door until 2005, and it is unlikely that the company has ever run at a profit. An SEC filing by Oracle regarding the acquisition said Pillar owes Ellision “and his affiliates” about $544 million for loans and interest.
During the Oracle call last week, Ellison said most tech companies on the market now are priced too high to acquire.
“I think we’re able to grow through acquisitions when they’re attractively priced and they make sense,” he said. “They are by and large not attractively priced now and don’t make sense, so we’re not doing them. If these assets are wildly overpriced, we can’t make a good business case for buying them. Instead, we can focus our energies on organic growth.”
Oracle will pay nothing up front for PIllar but might have to pay Ellison and Pillar stockholders if Oracle makes a profit from the Pillar products over a three-year period from the date the deal closes.
In a blog today on Pillar’s website, Pillar CEO Mike Workman said he and PIllar president Nancy Holleran will join Oracle. “Pillar is now a critical component to the Oracle storage strategy,” Workman wrote.
IDC today released the results of its annual EMC-sponsored Digital Universe study, which confirms what storage professionals see first-hand every day: data keeps growing unchecked and resources to manage it aren’t growing nearly as fast.
IDC forecasts that 1.8 zettabytes of data will be created and replicated this year – enough to fill 200 billion two-hour high-definition movies, 57.5 billion 32GB Apple iPads or the amount of storage required for 215 million high-resolution MRI scans per person per day.
In other words, a really lot of data, and it’s doubling every two years according to IDC’s numbers. And metadata is growing twice as fast as the digital universe.
Looking farther out, IDC forecasts that by 2020 IT departments will have 10 times as many virtual and physical servers, 50 times as much information, and 75 times the number of files or containers that encapsulate information than they do today.
And there will be 1.5 times the number of IT professionals to manage it all.
As you would expect, EMC global marketing CTO Chuck Hollis hit on the “big data” theme in discussing the results, but also suggested the findings could serve as a wakeup call to change the way people manage data.
“I would use this as evidence to go to senior management and say ‘We need a new game plan here,’” Hollis said. “Simply expanding five percent year-over-year on storage costs, taking the machines they have and tuning them up – that’s not going to keep up. I meet a lot of storage people who think they’re like the people with their fingers in the dikes, the water keeps coming and they’re running out of fingers and toes. Maybe it’s time to think about this problem differently.”
Hollis said “a lot of people are looking at this as an opportunity instead of a problem,” and those people are what EMC refers to as the “big data crowd.” They consist largely of media and entertainment companies and researchers who use data to make money for their employers.
“There are actually two kinds of IT organizations we see often in a big company,” Hollis said. “One is the traditional IT guys who deal with shared services, e-mail, Oracle and things like that. The big data crowd is usually a separate IT structure, usually researchers or business guys who have an idea and they handcraft the environment in such as way that makes the money or provides the value they want. The technology is different, the organization is different, and the thinking is different. At what point does this big data IT start looking like mainstream IT? Certainly not this year, but if this data growth keeps going, in three or four years it will be a lot more complex.”
IDC group vice president for storage Dave Reinsel said data growth is fueled partly by the low cost of disk. But he agrees with Hollis that organizations need to take a different look at how they deal with the data.
“We’ve made it dirt cheap to store,” he said. “If costs were going up like gasoline, people might change their behavior. But storage cost per gig is going down every year, so people have more. But data centers aren’t cheap to run. You have to justify building another data center. We’re getting to the point where we need to enable companies to extract the value out of that information.”
So far, Reinsel said, cloud storage isn’t playing much of a role in storing that information. Today, all cloud computing accounts for less than 2% of IT spending.
“Only 20% of information will be touching the public cloud by 2015,” Reinsel said. “People aren’t just jumping to public clouds. Hybrid clouds are out there and social networks are driving growth to public clouds, but there are still security concerns.”
Following a year of large storage acquisitions, it looks like 2011 might be more IPO-friendly for storage vendors.
Two weeks after solid-state storage vendor Fusion-io went public, clustered NAS provider BlueArc Friday registered with the Securites and Exchange Commission (SEC) for a public offering. Nexsan already has an IPO filing on the books and SAN vendor Xiotech’s CEO Alan Atkinson said he is looking to follow Fusion-io’s lead and go public.
BlueArc has gone this far before. It filed for an IPO in 2007 but never followed through because of poor market conditions.
BlueArc, which benefits from an OEM deal with Hitachi Data Systems, has never had a profitable quarter and has lost a total of $230.3 million since it began shipping its storage systems in 2001. Its annual revenue was $74.2 million in 2008, $65.9 million in 2009 and $85.6 million last year, and it lost $19.6 million, $15.8 million and $9.4 million over those years.
In the three months that ended April 30, 2011, BlueArc had revenue of $24.7 million and lost $4.3 million.
The BlueArc filing said the vendor has more than 750 customers with more than 2,000 of its systems deployed.
SAN vendor HDS sells BlueArc’s SiliconFS file system with its storage arrays to give HDS platforms NAS capability. HDS accounted for 41% of BlueArc revenues last year and 45% of its revenues for the quarter ending April 30. BlueArc’s filing said its contract with HDS must be renewed every year. However, Judging from public statements HDS has made, it is happy with the BlueArc relationship.
BlueArc’s filing said it hoped to raise up to $100 million in the IPO. That’s small change compared to some of the storage transactions over the past 12 months. EMC acquired BlueArc competitor Isilon for $2.25 billion last year. Also over the past year, Hewlett-Packard bought 3PAR for $2.35 billion, Dell acquired Compellent for $820 million, and NetApp picked up LSI’s Engenio storage division for $480 million.
The people who run Xiotech are closely watching Fusion-io these days.
That’s because the rollout of its Hybrid ISE solid state storage system this month has increasingly brought Xiotech into competition with PCIe flash card vendor Fusion-io. Xiotech is also looking to go public eventually, and Fusion-io’s IPO this month raised $237 million.
Xiotech CEO Alan Atkinson said Xiotech will ship every unit of Hybrid ISE it could build this quarter, although he didn’t say how many units were built. “This will be the most successful product launch in Xiotech history, and the first of several products on our roadmap in quick succession that will work together,” he said.
Atkinson said part of the success of Hybrid ISE is due to the awareness of the flash market that Fusion-io created with its products and the attention its IPO created.
Xiotech takes a different approach to SSDs than most storage array vendors. Instead of using SSD as cache or plugging SSDs into traditional arrays, Xiotech puts a set amount of SSD capacity along with hard drives in its storage bricks. Each brick has 20 hard drives and 20 SSDs to provide 14.4 TB of usable capacity, and uses what Xiotech calls Continuous Adaptive Data Placement to move data between hard drives and multi-level cell (MLC) SSDs to optimize I/O performance.
Atkinson said Hybrid ISE is shipping mostly into new markets for Xiotech. The new customer base includes Fortune 500 firms, particularly financial services companies looking to accelerate database performance. “That’s not shocking,” Atkinson said. “That’s where Fusion-io is selling, and that’s where the SSD market seems to be.”
Like Fusion-io’s products, Hybrid ISE appeals more to the people who manage applications than traditional storage admins. Along with Oracle databases, SSDs in storage are a good fit for virtual desktop infrastructures (VDIs).
Atkinson said Xiotech’s advantage is that Hybrid ISE is easier to set up and manage than PCIe cards. “Fusion-io goes to the apps guys and says ‘We can make your stuff look really fast.’ And it’s true,” he said. “But the administration of that is pretty difficult. They have to take a small LUN, open the servers up, put a card in, and roll their own DR solution because there’s no built in replication that looks like disk. And they have 800 gig as a target. That means they have to re-architect things.”
The storage vendor landscape has been re-architected the past few years as the most successful smaller companies have been gobbled up by the big guys. 3PAR, Compellent, DataDomain, EqualLogic, and Isilon all started around the same time as Xiotech but Xiotech is still on its own while the others have been absorbed by Hewlett-Packard, Dell and EMC. And most of those deals have been for billions of dollars.
Atkinson, who sold software vendor WysDM to EMC in 2008 before joining Xiotech, said it’s good to be among the few smaller storage system companies left standing.
“For a private company, those types of acquisitions raise your profile,” he said. “It makes it easier for us to look at a public offering, which is the path we’re on. There’s a dearth of companies in that space and storage has demonstrated itself to be hot. There’s a real appetite in the [financial] community for storage companies.”
During many of my discussions with IT managers and directors who would be classified in the mid-tier enterprise space (over 1,000 employees), it has become clear that few have deployed storage systems using solid state drives (SSD) with internal tiering. This surprises me given the performance improvements gained by using SSDs as the highest performing tier.
When I ask why they have not installed these types of tiering storage systems, I get some interesting responses:
• A belief that SSDs would be too expensive for their environment;
• They were unaware of tools to figure out how much capacity should be in SSDs to maximize performance; and
• Tiering storage systems were believed to be solely high-end enterprise solutions.
The value of tiering storage to improve performance while automatically managing the movement of data based on patterns of access has been demonstrated, and there are case studies available from vendors. The performance improvement is measurable and most vendors with offerings have tiering monitoring and reporting tools that can show the positive effects.
Additionally, analysis tools can help determine the correct amount of storage capacity in SSDs to maximize performance based on the workloads. Most analysis has shown that on average about 4% of capacity in SSDs can provide the greatest gain.
Using storage tiering as an immediate improvement for storage demands in server virtualization environments is a great benefit in the mid-tier. In this market, server virtualization is moving to primary business applications and the performance of storage can become a critical bottleneck. Articles on storage tiering are available at the Evaluator Group web site.
The real problem here is the ineffective marketing messages from the storage vendors. The message about the value, costs, and tools is not being received by the people that need to hear it. Some of the questions I ask include where the IT directors and managers get their information. The vendors looking to sell to the mid-tier need to be more targeted with their message and use a different approach than with the enterprise data center customer. The presentation of the information also needs to be in the context of what the mid-tier IT person is trying to address with a storage purchase. The value is real and demonstrated but the vendors have not made this point.
For any vendor who wants to accelerate successes with tiered storage systems in the mid-tier environment, a focused, special effort is required. Otherwise, another vendor may take that business.
(Randy Kerns is Senior Strategist at Evaluator Group, an IT analyst firm).
EMC CEO Joe Tucci isn’t publicity shy, but he almost certainly wishes he could have avoided his front-page exposure in the Wall Street Journal this week.
The Journal Thursday led a story on corporate jet abuse with Tucci’s use of EMC jets. The Journal claimed EMC jets made 393 trips over four years to areas where Tucci has vacation homes – Cape Cod, Mass., the New Jersey shore and the Florida keys. That comes to more than 98 flights a year to those sites.
The newspaper estimated the cost of EMC’s flights to those airports at $3.1 million, while noting that EMC puts the cost at $664,079.
A follow-up story in the Boston Herald today quotes an email from an EMC shareholder who took exception to Tucci’s jet-setting on the corporate dime:
“In exceptional cases, we sometimes speak out with other share owners about executive excesses,” wrote Clark McKinley, spokesman for the California Public Employees’ Retirement System, in an e-mail. “As far as I know now, this is one of them.”
The Herald story said CPERS owns 5.4 million shares of EMC stock.
Tucci, 63, ranks 15th on Forbes’ 2011 corporate compensation list with $31.63 million in salary, bonuses and stock gains. He has been paid $86.79 million over the past five years, according to Forbes.
In the high performance computing (HPC) world, Lustre serves as a clustered file system meeting needs for extremely large numbers of files and extremely large file sizes. The problem is, Lustre has been used primarily for “build your own” storage systems with questionable support.
Xyratex has moved to solve this problem for HPC customers with the ClusterStor 3000. The ClusterStor 3000 is a scale-out Lustre storage system that can support tens of petabytes in capacity and from 2.5 GBps to 1TBps throughput in performance. The ClusterStor 3000 ships with Lustre integrated, as well as full support and an integrated management tool.
ClusterStor 3000 customers will have a fully supported, high performance storage system that uses the Lustre file system. The ClusterStor system also includes ClusterStor Manager, a single administrative interface that simplifies configuration and management tasks instead of requiring admins to manage individual element as in previous Lustre deployments.
Xyratex last year signaled its commitment to Lustre when it acquired ClusterStor, a startup including founding members of the Lustre team Peter Braam and Peter Bojanic. Braam and Bojanic remain at Xyratex.
Xyratex has a long history of delivering components and systems to OEMs for a wide range of products. Its OEM partners include NetApp, IBM, Dell and EMC. Offering the ClusterStor 3000 is a major step for Xyratex because it matches a need in the market for HPC storage with a complete system based on a fully supported version of Lustre.
Like Xyratex’s other products, ClusterStor 3000 will be sold through OEM partners. None have signed on yet, but Xyratex executives expect it to hit the market late this year. Xyratex positions ClusterStor as competitive to Lustre-based systems from DataDirect Networks and NetApp’s Engenio division.
(Randy Kerns is Senior Strategist at Evaluator Group, an IT analyst firm).
Understanding storage technology can be difficult enough without vendors adding to the problem with odd product positioning. Yet vendors often make things worse when talking to customers or prospective customers.
I recently had conversations with two IT professionals that brought to light this issue. The first one was looking at backup software for an optimization project. The IT pro I talked to wondered why one vendor had two products with many common characteristics and if both would be continued. He was concerned that one product would be dropped or put into maintenance mode with no additional upgrades. The concern was justified because he intended the backup software to have a long usage period in the data center.
The salesman was not particularly enlightening about the vendor’s long-term plan, and no public information was available to clear up the situation. In this case, acquisitions had led to the two offerings. The vendor had different messages for each product and no message about the two in a combined plan. This reminded me of a Dilbert cartoon strip called “Battling Business Units” showing internally competing businesses that did not play together.
The second example involved the purchase of a disk storage system. In this case, several products were being considered to bring to a short list for final evaluation. One vendor had two products that might satisfy this customer’s needs, but there was much overlap between the two systems. This IT pro wondered how a vendor could continue both products. Investing in a storage system with training and operational procedures could be compromised if he bought the system that would eventually be dropped.
Again, acquisition had led to the vendor having two products and the messaging around positioning and continuation was not clear enough to remove the concern. Maybe it was another case of Battling Business Units. In any case, there was not enough coordination between the units to notice these obvious questions.
While working with IT in evaluations (see Evaluator Group for evaluation guides), I find that the type of information IT pros need from vendors is often missing or conflicting. This requires them to spend time on issues besides the product and the underlying technology.
Some vendors may embrace the Battling Business Units scenario and the internal competition it brings out with the philosophy that the best team will win. But it is not in the best interest of an IT customer making decisions.
By the way, Dilbert is not really a comic strip. It’s a documentary.
(Randy Kerns is Senior Strategist at Evaluator Group, an IT analyst firm).
Quantum acquired startup Pancetera Software today, giving it virtual server backup for its DXi data deduplication family right away. In the long run, Pancetera can also add intelligent storage management for virtual environments to Quantum’s StorNext file system.
Quantum paid $12 million for Pancetera, which came out of stealth last August with its Pancetera Unite virtual appliance designed to optimize virtual machine backup. It added SmartMotion software in April, enabling Unite to push data from virtual machines directly to any NAS target without requiring staging servers with dedicated backup software.
By owning the technology instead of forging a partnership, Quantum is looking to develop new products for protecting and managing storage connected to servers running VMware.
“This gives us immediate value for DXi and virtual environments, and it will allow us to have unique roadmap items with DXi,” Quantum CEO Jon Gacek said. “The reason we went with an acquisition instead of an OEM deal is we can combine this with StorNext to develop solutions to manage storage – not just backup, but storage – in virtual environments.”
Gacek said Quantum will immediately offer Pancetera software with DXi systems sold for virtual environments. “We’re in deals now where we know that software will make the difference,” he said.
He said Quantum will also sell Pancetera to existing DXi customers, but he’s not sure if it will make Pancetera software available as a widescale standalone product. “We haven’t decided if we want to enable its value to our competitors’ customers,” he said. He said he expects Pancetera technology with StorNext to hit the market in an appliance for virtual data in 2012.
“Virtualization creates a lot of unstructured data,” Gacek said. “StorNext with Pancetera gives us the ability to get inside of a VMDK, and you can imagine some of the things we might do.”
Quantum tried attacking the VMware backup problem with OEM partner PHD, licensing its esXpress product in 2009. But that relationship fell apart last year. Gacek said the OEM deal didn’t give Quantum enough control over the technology.
Quantum is hiring most Pancetera employees, including founders Mitch Haile (CTO) and Greg Wade (VP of engineering) and CEO Herik Rosendahl.
“The issue for them was, they were going to be hard-pressed to convince companies to buy that kind of software from a startup,” Gacek said. “But the software is complete. We like how it can be supported and it fits well with what we’re doing with DXi.”
VMware is the only hypervisor that Pancetera supports. A Quantum spokesperson said there has been no decision yet on whether it will expand it to other hypervisor platforms.
Gacek, who replaced Rick Belluzzo as CEO in April, said the acquisition and the hire of Ted Stinson as senior VP of worldwide sales Monday shows “we’re in growth mode and we’ll be aggressive about making change.”