Steven Young, a site reliability engineer at Orbitz, is presenting at CouchConf on Friday afternoon about making the switch from Coherence to Couchbase, and he shared some details with me in advance of his presentation. Essentially, Orbitz moved some very important systems — including its Hotel Rate Cluster, which caches all the relevant information on hotels and rooms and delivers it to customers — over to Couchbase and was able to achieve serious improvements in cost and performance while slashing its footprint:

• Four hundred nodes have been reduced to 70 nodes that handles about 250 million objects and 1.2TB of data.
• In-memory cache size reduced to 55GB from 620GB.
• The company is saving more than $2 million a year in licensing, maintenance and operations costs.
• Orbitz can now reliably scale clusters across data centers.
• Latency has improved significantly, and multiple-times-per-week failures have ended.
• Orbitz can now roll out changes to its web applications without bringing them down.

All told, Young said, Couchbase is just faster, easier, more efficient and more scalable. “We have not encountered any kind of negative tradeoff, everything has been positive at this point,” he said.

The point-of-sale cluster went form 32 nodes (outside) to 10 (inside). Orbitz is testing a reduction to six nodes.

But like Disney, which also recently made the switch to a collection of open source technologies, Orbitz isn’t playing around when it comes to making sure its systems are running optimially. While a handful of lower-value web applications are still running Membase or memcached, most are now running Couchbase’s Enterprise Edition, Young said. Free versions of open source products work for areas such as testing and development — heck, Orbitz even built and open sourced its own reporting tool called Graphite — but big companies have to choose carefully where to use them.

Because hotels are the key to any travel business, Orbitz is perfectly willing to pay if it means those systems are displaying fast, accurate information to customers. “We like to make sure we’re getting the most for our cache in every sense of the word,” Young joked (bonus points for spotting the pun).

In fact, Oracle’s top-of-the-line and very expensive Exadata appliance still handles some very important work for Orbitz. Whereas Couchbase stores what’s relevant at any given time, Exadata stores a year’s worth of hotel rate data that the company needs to keep for compliance and historical analysis reasons. “It’s not feasible for us to store all of that in Couchbase,” Young said. “… It really depends on the strategy we have.”

Feature image courtesy of Shutterstock user concept w.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Defining Hadoop: the Players, Technologies and Challenges of 2011
• A near-term outlook for big data
• The Case for Increased M&A in 2011: Actions and Outlooks

For Violin the news is (what else?) more money. The company is expanding the $50 million Series D round it closed in late March to $80 million and is making room for new strategic investor GE Asset Management. The previous $50 million was at an $800 million valuation, although Violin CEO Don Basile told me then he expects the company is worth billions should it go public later this year.

A couple commenters noted in my post on the EMC-XtremIO rumors that Violin is all about performance, and that such a focus won’t necessarily win the day for flash-storage companies hoping to become standard fare within enterprise data centers. Whether that’s a fair assessment of Violin is open for debate, the commenters’ general point about quality of service, virtualization integration and other enterprise-grade features as the true differentiators is well taken. If all car buyers cared about was speed, luxury automobiles would resemble racecars more than dens. Still, barring a colossal collapse, Violin’s piles of cash, huge valuation and laundry list of strategic investors suggest it has the goods to find itself among the industry leaders for a long time.

For Fusion-io, which went public itself last year, and has been flying high since, its momentum comes in the form of early support behind its recently released ioMemory software development kit. Fusion-io and Violin are architecturally quite different — the former being a server-side PCIe component and the latter being a network-attached storage array — but Chris Mellor at The Register offers an insightful take on how certain deployment scenarios are blurring the line between the two companies.

Specifically, he writes, deployments of new Fusion-io SDK partner Couchbase and new Violin partner SAP (with its HANA in-memory database) end up looking remarkably similar in theory. Either way, hot application data resides in-memory (DRAM), while everything else resides on flash. Couchbase on Fusion-io is more like a motorcycle with the passenger holding onto the drivers waist, while SAP on Violin is more like a motorcycle with the passenger riding in a sidecar, but they’re both accomplishing the same goal.

If Violin and Fusion-io do start competing more directly, we’ll be in for a wild ride. What seems certain, however, is that it’s just a matter of when, not if, flash storage catches on in a major way. With more money and more software-vendor support always flowing into the market, there will soon be precious few reasons not to give the technologies a serious look.

Image courtesy of Shutterstock user fotomak.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Infrastructure Q4: Big data gets bigger and SaaS startups shine
• Infrastructure Q2: Big data and PaaS gain more momentum
• Infrastructure Q1: Cloud and big data woo enterprises

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• A near-term outlook for big data
• Defining Hadoop: the Players, Technologies and Challenges of 2011
• Infrastructure Q2: Big data and PaaS gain more momentum

Build or buy?

What stands out most when looking at Zynga’s gaming infrastructure is that Zynga prefers to buy software or use commercial open-source software rather than build its own. Its zCloud is built upon cloud computing software by Cloud.com (now part of Citrix), its analytics platform is primarily a Vertica database (now part of HP) and its application servers run the Couchbase NoSQL database. And although it has significantly lessened its reliance on the public cloud, Zynga still runs 20 percent of its gaming workload on Amazon Web Services.

Facebook? It runs its own Hadoop distribution, created the Cassandra database and has built countless tools for automating processes and speeding applications. It has scaled MySQL to extremes heretofore thought impossible. Heck, it even builds its own servers and storage as well as designing its own data centers from the ground up. It’s all very impressive, but is it money well spent?

Facebook spends so many man-hours building new software and hardware because it can tune them to the platform’s unique demands and because it can save untold millions on new gear, licenses and energy costs annually as the the infrastructure scales to handle nearly a billion users.

So why buy?

When I asked Zynga CTO Allan Leinwand about his company’s buy-over-build mentality, he told me it was a matter of what’s strategically important. Zynga focuses on deploying infrastructure that “satisfies the needs of our business,” he said, and Zynga’s core business is delivering an engaging gaming experience. “The most important thing for us is building infrastructure that runs our games in the most effective way,” he said, adding that he’d run Zynga on his iPhone if that provided the best user experience.

And it’s not as if Zynga’s zCloud didn’t require some heavy-duty engineering work. As Leinwand explained, the company wants its platform “to be available to everyone on the planet all the time,” and getting to that point meant lots of fine-tuning across the computing, storage and network infrastructure. It has gotten to a point where one of Zynga’s own servers can handle three times the load of a comparable machine image on AWS. Actually, being able to effectively utilize the public cloud for spikes in workload is itself an impressive feat for a company of Zynga’s size.

Fair enough. In fact, buying over building is an approach I’ve championed in the past (although one that I’ve softened on), especially for web startups. But Zynga is a billion-dollar company. If it grows as it wants to — and as its investors no doubt expect it to — the costs of license costs and limitations of commercial software could get cumbersome.

The numbers tell the story

Zynga’s latest financial results illustrate why the company is evolving into a platform provider, but they also suggest the company needs to nail that transition. Yes, the company earned just more than a billion dollars in 2011, but it ended up actually losing $405 million on the year. About $238 million (or 22 percent) of Zynga’s earnings went toward building out its infrastructure. Roughly $445 million went to Facebook as the cost of doing business on its platform.

Facebook, on the other hand, earned more than $3.7 billion in revenue in 2011 and netted a respectable $1.7 billion before taxes. It spent $606 million (or 16 percent) of its earnings on building out its infrastructure. And while Zynga said it expects to spend “in the range of $140 million to $160 million” on capital expenditures in 2012, Facebook is planning more than 10 times that amount as it builds out its data centers around the world.

Assuming it can bring its hundreds of million of users — as well as a vibrant developer ecosystem — to Zynga.com, perhaps Zynga can get back in the black. But what if it grows too fast (or too slow), or what if it wants to get a more balanced revenue stream between in-game purchases (currently about 90 percent) and advertising (currently about 10 percent)? Advertising revenue, after all, is the lifeblood of most web-platform companies (subscription required).

Facebook’s work on its own infrastructure technologies has enabled it to keep scale right along with its user base and their increased activity on the site. Facebook (as well as other web platforms such as Yahoo and Google) operates a massive Hadoop cluster that helps the company build features and target advertising based on what it’s able to learn about its users. Zynga has to hope its off-the-shelf pieces can scale — and do so affordably; if it needs to expand its analytics efforts, it might need to complement the Vertica system with a significant Hadoop deployment.

Being a major web platform is serious business on the infrastructure side, something Zynga no doubt understands. And while its current technology infrastructure is impressive and innovative, history doesn’t seem to be on Zynga’s side if it wants to grow into something as big as Facebook with its current system. Leinwand seems to get this. Zynga is always looking at new technologies and new ways to innovate, he said, because “forever’s an awful long time” to commit to anything on the web.

Feature image courtesy of Flickr user randomduck.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• The capex connection: Why we pay for privacy on the Web
• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight
• A field guide to cloud computing: current trends, future opportunities

Where Riak comes is as the database that stores Bump user information and transfers it immediately to other users when they physically bump their phones. Here’s how Basho described its role in a press release:

Bump stores elements of current and past conversations, so communication history is readily accessible to users, enhancing the stickiness and user satisfaction of Bump.

… Riak ensures that the Bump application can be continually fed with information without the worry of a system fail. Bump stores 600-800 million pieces of structural data in Riak, including photos, chats and contact cards.

Mobile apps are actually a pretty big sweet spot for NoSQL databases because of the easily accessible and widely distributed natures of mobile apps. Lots of users, lots of semi-structured data and users’ expectations of immediate satisfaction mean mobile apps need fast, scalable and flexible databases.

Basho counts a number of mobile app developers among its users, and there a number of mobile apps, including free-SMS app Viber, using the MongoDB database. Couchbase has put a focus on syncing mobile-app data between devices and the cloud (although its has changed its strategy a bit since abandoning CouchDB earlier this year).

We’ll be talking a lot about both NoSQL and mobile data at our Structure:Data conference next month in New York, which features 10gen CEO Dwight Merriman and Wordnik Co-Founder Tony Tam, as well as a handful of experts on storing and analyzing mobile data.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• The fourth quarter of 2012 in cloud
• How devops can reduce cycle times
• Cloud and data third-quarter 2012

It’s not surprising that respondents had a NoSQL bent given that Couchbase recruited them from Twitter (using the NoSQL hashtag), InfoQ as well as the company’s own database. Still, some of the results were interesting.

NoSQL (or not only SQL) databases, which do not rely on the SQL query language commonly used in relational databases from Oracle,  Microsoft, and IBM, got their start in webscale companies like Google and Yahoo that needed their ability to scale out across multiple servers. MongoDB, Cassandra are other NoSQL databases.

Of those planning to make the transition, the biggest motivation was disdain for the inflexible schemas that are part-and-parcel of SQL databases, according to the survey. Nearly half (49 percent) of those surveyed said schemas drove them to NoSQL. The second biggest factor, cited by 35 percent, was the inability of relational databases to handle scale-out data.

One of NoSQL’s biggest draws is that users don’t have to scope out their data fields — name, phone number, state — in advance and then be held captive to those schemas.

It was striking that the language cited by most of NoSQL-loving respondents was Java, followed closely by PHP, then C#.  ”That could be surprising that the favorite was not Ruby or the bleeding-edge funky languages,” said James Phillips, Couchbase co-founder and SVP of products.

NoSQL is definitely gaining traction but the pure-play NoSQL players will see increasing competition from the RDBMS world as well. Virtually all the RDBMS players including Oracle are coming out with their own NoSQL plays.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• A near-term outlook for big data
• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight
• Infrastructure Overview, Q2 2010

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Infrastructure Q2: Big data and PaaS gain more momentum
• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight
• Big Data, ARM and Legal Troubles Transformed Infrastructure in Q4

CouchDB, an Apache Software Foundation project, is a fan favorite among many NoSQL developers who laud its local replication and support for many devices which suit it for mobile app development. But Katz said the time has come for change, and his rationale appears to stem from concerns he has about the open-source development model itself. According to his blog:

With Apache CouchDB, much of the focus has been around creating a consensus based, developer community that helps govern and move the project forward. Apache has done, and is doing a good job of that. But for us, it’s no longer enough. CouchDB was something I created because I thought an easy to use, peer based, replicating document store was something the world would find useful. And it proved a lot of the ideas were possible and useful and it’s been successful beyond my wildest ambitions. But if I had it all to do again, I’d do many things different.

Going forward, Katz will focus on Couchbase Server 2.0, and rewrite more of its core services in C/C++ rather than Erlang.  ”Most of the performance sensitive portions [of CouchBase] will be moving to over C code. Erlang is still a great language, but when you need top performance and low level control, C is hard to beat,” Katz wrote.

He said the new product will build on previous Apache CouchDB and Membase projects, but his team will jettison what doesn’t work and focus on a design for “mission critical deployment and mobile integration.” Katz said that work can move faster and be more responsive to customers than a “community based project.”

Hello Couchbase 2.0, bye-bye Couchbase Single Server

News of Katz’ decision came as Couchbase discontinued CouchDB-based Couchbase Single Server, citing self-inflicted market confusion. According to a company post:

In addition to Membase Server and our new mobile products we also offered Couchbase Single Server which was a packaged “distribution” of Apache CouchDB. On top of that we began releasing developer previews of Couchbase Server 2.0, which incorporated CouchDB technology into Membase Server – but this product was not compatible with Couchbase Single Server (or CouchDB). If you are confused just reading this you get the point – and so do we.

News of Katz’ transition sparked a lively debate on Hacker News. Some defended CouchDB as a valid NoSQL choice while others said business users need more professional support than was available with CouchDB. Some commenters slammed Katz for disrespecting the project that brought him fame. Several of the participants worried that Couchbase Server represents a “fork” of the CouchDB code base.

One commenter, Rubyrescue, wrote:

“This is business, pure and simple. I had a long, LONG series of emails and calls with Couchbase about commercial support for Couch. We have some big production apps on it. When it got down to time to pay for support, they told me (this is 2 months ago) in a rare and unusual bit of candor, that they were going to drop Couch in less than six months, so did I want to buy commercial support for just six months? I told them not only do I not want commercial support, but I just got so freaked out I would not recommend Couch for future projects to clients, because it was obvious that internally the team had moved on.”

What’s the migration plan?

Others complained that there doesn’t appear to be an easy CouchDB-to-Couchbase Server migration path or they would consider doing that.

The tension between open source and commercial software development is a persistent undercurrent in the tech arena. Open source proponents love having a large, democratic community contribute expertise both on actual coding and for support. Others say that’s all fine and good, but real businesses need a real, professional and paid support — an actual number to call for help. In this case it appears that the open source model took one on the chin.

Photo courtesy of Flickr user quinn.anya

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight
• Defining Hadoop: the Players, Technologies and Challenges of 2011
• How devops can reduce cycle times

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Cloud computing infrastructure: 2012 and beyond
• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight
• Infrastructure Q2: Big data and PaaS gain more momentum

Johri has a long resume. He is a director at Dinghana, and served as a top executive at several Silicon Valley companies, including at Hewlett-Packard where he served as a VP for both cloud computing and enterprise strategies. There, he helped drive HP’s $4.5 billion acquisition of Mercury Interactive in 2006 and its $1.4 billion buyout of Opsware the following year. Those acquisitions are now a key part of HP’s enterprise software and systems management business.

Before that, Johri founded Oblix, an identity management specialist, subsequently acquired by Oracle. He also had stints at Silicon Graphics and Gemini Consulting, and actually was an entrepreneur in residence at Mayfield from 2001 to 2003.

In a statement, Johri said, “I look forward to partnering with Mayfield to uncover big opportunities such as identifying how big data analytics can be targeted at specific industry segments like telecom and financial services.”

Recent research indicates some VCs and private equity firms aren’t confident enterprise software as a category has huge growth potential. Moves like this one shows at least some VCs believe that to be false.

Menlo Park, Calif.-based Mayfield has more than $2.7 billion under management and typically invests in early-stage companies. Mayfield currently has investments in Couchbase, Appcelerator, Marketo and StorSimple, all of which show an enterprise software focus. In the past, it has taken stakes in 3PAR, now owned by HP, LSI , and Riverbed Technology .

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• A near-term outlook for big data
• Infrastructure Q2: Big data and PaaS gain more momentum
• The Case for Increased M&A in 2011: Actions and Outlooks