A: McDipper, a Facebook-built implementation of the popular memcached key-value store designed to run on flash memory rather than pricier DRAM.

Memcached, for the unfamiliar, is an open-source key-value store that caches frequently accessed data in memory so applications can access and serve it faster than if it were stored on hard disks. It’s a very popular component of many web applications stacks, including at Facebook where the company runs thousands of memcached servers to power its various applications.

But DRAM is expensive, especially when you get to Facebook’s scale, and not all applications deserve that kind of performance. So, according to a Facebook Engineering post on Tuesday, the company designed McDipper to handle “working sets that had very large footprints but moderate to low request rates. … Compared with memory, flash provides up to 20 times the capacity per server and still supports tens of thousands of operations per second.”

Facebook has deployed McDipper for a handful of these workloads, the blog states, and has “reduced the total number of deployed servers in some pools by as much as 90% while still delivering more than 90% of get responses with sub-millisecond latencies.” It has been part of Facebook’s photo infrastructure for about a year and serves 150 gigabits of data per second — or “about one library of congress (10 TB) every 10 minutes” — over Facebook’s content-delivery network.

How McDipper stores data

This is the same logic that drove Facebook to undertake its cold storage engineering effort for even more infrequently accessed data, which aims to find a middle ground between the inefficiency and latency of hard disks and the high cost of flash storage. To meet that goal, the company is getting creative by considering everything from lower-performance flash to Blu-ray — pretty much anything but tape — VP of Engineering Jay Parikh told me in January.

Building a tool like McDipper is the just the tip of the iceberg, though, when it comes to managing the cost and efficiency of infrastructure at large web companies such as Facebook. On Tuesday, eBay released its Digital Service Efficiency report that lays out a methodology for assessing the effect that infrastructure (more than 52,000 servers in eBay’s case; Facebook has even more) has on larger corporate goals such as clean energy and the bottom line.

And later this month at our Structure: Data conference, data center executives from Facebook, Microsoft and Goldman Sachs will take the stage to discuss how smart analytics help them plan to meet capacity needs while keeping costs in check.

Feature image is Facebook’s new all-flash Dragonstone server design.

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

• Migrating media applications to the private cloud: best practices for businesses
• How tomorrow’s mobile-centric data centers will look
• Takeaways from the second quarter in cloud and data

EMC’s new XtremSF PCIe cards come in a few sizes. Enterprise multi-level-cell models with 550GB and a 2.2TB capacities are now available, and 700GB and 1.4TB models will come in the second quarter of the year. More sizes will follow. Last year EMC introduced two PCIe cards — 350GB and 700GB — under the name VFCache. Those two have joined the XtremSF line alongside the four new cards, said Barry Ader, general manager of EMC’s flash business unit.

EMC and others in the data center storage market have turned nearly 180 degrees from where they were just a few years ago, when they decried in-server flash memory from such companies as Fusion-io and sang the praises of separate flash memory arrays instead. Server-side storage eliminates the bottleneck between the processor and storage in separate boxes, decreasing latency.

It’s also a reaction to a fast-growing market, with webscale companies such as Facebook spending for fast-acting server-side flash storage. Fusion-io reported a 43 percent gain in year-to-year revenue for the fourth quarter of 2012, coming in at $120.5 million, according to figures on file with U.S. Securities and Exchange Commission.

With more products coming into the flash PCIe market, prices could fall further, accelerating enterprise adoption. Then again, companies could still find a way to compete in this new market by offering different capabilities, which mean other companies would need to jump in before prices race to the bottom.

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

• Will Storage Go the Way of The Server?
• Infrastructure Q4: Big data gets bigger and SaaS startups shine
• Infrastructure Q3: OpenStack and flash step into the spotlight

As the company branches out with four PCIe cards, with memory capacity ranging from 1.37 terabytes to 11 terabytes, it will bump up against several competitors, including Fusion-io, LSI and Virident.

The Violin Memory cards are bootable, which can save time and minimize frustration. They also use less air flow than other flash memory cards, so customers can pack more cards onto servers. The lowest-capacity card costs $3 per gigabyte, and the price goes up to $6 per gigabyte for the others.

“We believe this (1.37 terabyte) card, with its price-performance level density, will allow the industry to start a broad adoption over the next several years,” said Don Basile, Violin Memory’s CEO.

The company can rest assured of a market for the new cards. Toshiba, already a major flash memory vendor for consumer products, will have licensing and distribution rights for the new Violin Memory intellectual property.

Violin Memory is planning to go public at the beginning of May, All Things D reported. Basile declined to discuss the timing of the product launch in relation to the IPO.

The flash storage market is nothing if not active. Last May, EMC acquired XtremIO; two months later, IBM bought Texas Memory Systems, and Pure Storage said it had secured a $40 million Series D investment. Just two weeks ago, flash storage array vendor Skyera announced a $51.6 million Series B round of funding.

Widespread enterprise adoption of flash memory seems to be a matter of time. With its new products, Violin Memory appears to be in a better position to ride the market wave.

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

• Infrastructure Q1: Cloud and big data woo enterprises
• Amazon’s DynamoDB: rattling the cloud market
• Infrastructure Q2: Big data and PaaS gain more momentum

There’s certainly plenty of demand for the services. In 2009, CIO reported that more than 40 million Americans had tried online dating. In September 2012, 1 in 10 internet users frequented an online dating site, according to a December 2012 report from comScore.

Sites vary as to the times of year when traffic peaks. The number of unique visitors to eHarmony.com increases 45 percent on Valentine’s Day, and the boost continues until the end of the month, a spokeswoman wrote in an email.

The biggest day of the year for Match.com registrations isn’t Feb. 14; it’s actually Jan. 2. “Then we get another big spike after the Valentine’s Day holiday, so this weekend will be another spike,” a spokeswoman said. And for Zoosk.com, the peak comes on Dec. 26, while traffic is consistently heaviest in January.

How do engineers accommodate all the traffic and not sacrifice performance?

For eHarmony, it was a matter of scaling out infrastructure. “The systems over the years have been expanded to absorb large spikes to all the main areas and events on the site, such as posting photos, communication requests and the interactions with the mobile apps,” the company spokeswoman wrote.

Computerworld reported in 2009 that eHarmony had 4 terabytes of user information in storage for 20 million users. That comes from responses to the site’s Relationship Questionnaire. The spokeswoman did not immediately have current figures available.

Match.com was storing 70 terabytes of user data for more than 1 million subscribers when Microsoft published a case study on the dating service last March. Until May 2010, Match.com was updating user information on 110 Microsoft SQL Server servers across two data centers in the United States. In order to keep profile updates timely — less than two seconds — the company began distributing the updates across the servers, rather than update the entire dataset at once.

Valentine’s Day isn’t necessarily the high point of the year for Facebook. Jay Parikh, Facebook’s vice president of infrastructure engineering, cited Halloween as one of the highest times of year for photo uploading, as my colleague Stacey Higginbotham reported. When Facebook demand spikes, servers stocked with flash memory in the data center instead of hard disk drives and tapes ensure consistently high performance with a wide variety of data — and there’s plenty of room for storage, too. Facebook’s flash-only database servers, codenamed Dragonstone, feature 3.2 terabyte flash memory cards from Fusion-io. Flash memory might come in handy at dating sites’ data centers, too — the Dragonstone flash memory became part of the Open Compute Project last month.

Feature image courtesy of Shutterstock user 3Dstock.

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 Q3: OpenStack and flash step into the spotlight
• Infrastructure Q2: Big data and PaaS gain more momentum

GridIron’s appliance technology sits between a company’s servers and its storage systems, acting as a low-latency caching tier for high-value data. The appliance’s software constantly analyzes how applications are accessing data and then decides which pieces should reside where. Low-priority stuff might remain in the storage-area network (SAN), while high-priority stuff will move into DRAM or flash in the GridIron appliance that’s directly connected to the servers.

Flash arrays alone, like those Violin sells, already make databases and virtual-machine management significantly faster, so something like GridIron is really just icing on the cake. According to Violin CEO Don Basile, GridIron also has built a solid business in the worlds of video and video game production, and computer-aided design. These are areas where giving applications faster access to data means faster production so they can hit the shelves sooner.

However, there’s another way to look at the advent of flash as a first-class citizen in enterprise data centers, which is that it needs to address the needs of all applications rather than just those with extreme performance needs. As Facebook VP of Infrastructure Engineering Jay Parikh explained me recently, he thinks flash manufacturers (and by proxy, I assume, flash vendors such as Violin) actually need to broaden their product lines to address the needs of lower-performance applications. If the goal is to eliminate hard drives — which is Basile’s stated goal — Parikh thinks flash, Blu-ray and every other type of storage medium needs to include variations that resemble hard drives in terms of cost and reliability.

Don Basile

But rather than trying to work backward toward hard drives, Basile wants to make flash more durable and higher-capacity while still continuing to improve performance. In theory, if storage vendors can sell more enterprise-grade flash, its price tag will drop faster and flash manufacturers will invest more in enterprise flash technology as opposed to consumer flash technology that currently dictates what storage vendors have to work with. Violin actually has a strategic relationship with Toshiba that could help spur this shift, as might the growing number of flash-storage vendors including Fusion-io, Whiptail, EMC, Pure Storage and Nimbus Data.

“We do think they have a great vision over there at Facebook,” Basile said. However, he added, citing the gulf between Bill Gates’s rumored 640KB quote and Kingston’s new 512GB personal USB drive, “If we have speed, we don’t usually want to give it back.”

The deal is rumored to be worth between $200 million and $300 million.

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

• Flash memory: the continuing disruption of enterprise storage
• Infrastructure Q2: Big data and PaaS gain more momentum
• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight

Several of the investors have returned from an earlier $35 million round that the nine-year-old business had raised when it was pursuing a slightly different business, one that also relied on boosting the performance of memory in the data center. Riccardo Badalone, a co-founder and CEO of Diablo, said the company wouldn’t get too specific about its plans, but he did say that it is working on a silicon product that will marry the processor and flash memory to help boost performance. This is yet another evolution in bringing the memory and processing power closer together to speed up applications. Fusion-io does this, only its business was built on putting the memory on the PCI express card.

Badalone says he can bring it closer. That would help get more data to the processor and in turn let the processor take on more information. And in today’s web-based business or cloud environments, the more your processors run, the more money you make. That sentiment is why flash memory and ways to increase its performance are rapidly gaining interest –and why Fusion-io has done so well.

Diablo is borrowing a tactic from Fusion-io in that it is working with equipment makers to put its technology — called Memory Channel Storage — inside servers. Its previous product was qualified on Dell, HP and IBM systems, which is one reason that investors ponied up for the latest round even as Diablo was switching strategies. Badalone anticipates using similar partnerships with its chip technology, although he declined to name any of those potential partners or existing customers.

The company has 40 employees and plans to grow to 50 or 60 in the coming year, and Badalone is optimistic about the performance of the systems when it eventually hits the market in 2013. “We will be the first company in the world to marry RAM and Flash in the memory subsystem and once you do that the performance of the I/O will be such that people will lok at the numbers, and frankly, they will not believe it,” Badalone said.

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

• Takeaways from the second quarter in cloud and data
• A near-term outlook for big data
• Infrastructure Q3: OpenStack and flash step into the spotlight

It’s worth noting that the teardown is “preliminary in nature, account only for hardware and manufacturing costs and do not include other expenses such as software, licensing, royalties or other expenditures,” says the research firm. In light of that, IHS iSuppli suggests that the total bill of materials plus an $8 manufacturing cost for the 16 GB iPhone 5 is $207. Bump up the iPhone to 32 GB or 64 GB of memory and that number increases to $217 and $238 respectively; a vast difference from the phone’s full price to consumers, which are $649, $749 and $849, respectively.

Buying a subsidized handset still gets Apple its profit though; carriers pay hundreds of dollars per handset to Apple in return for voice and data plan commitments from the consumer.

One aspect of the cost breakdown that jumped out to me is that the 64 GB iPhone 5 model actually costs Apple $15 less in parts and production than last year’s phone. The other two models cost slightly more: Apple spends an additional $2 to create this year’s 32 GB phone and $11 more for the 16 GB version, per the IHS iSuppli figures.

Still, if Apple were to sell an equal amount of each model this year; it works out to a net gain: The $13 more to create the two lower models is more than offset by the savings on the high-end model. I’m not implying that Apple actually does sell an equal number across the model line, but it’s an interesting profit hedge scenario. And I suspect that more people — especially existing iPhone owners that are upgrading — gravitate towards the models with more storage capacity.

Where’s the big savings this year? The largest is Apple moving away from Samsung for its flash memory. SanDisk memory was found in the iPhone 5 torn down for the IHS iSuppli analysis, although the research firm says that Samsung, Hynix and Toshiba could be suppliers across the line too. This difference is huge by comparison to all other parts. Last year, the memory in Apple’s 64 GB iPhone 4S cost $76.80; this year, that part is $41.60.

Not only does this illustrate how well Apple manages its part supply prices, but also why it’s in a position to do so. Instead of multiple phone models with different sizes and varied components, the iPhone is generally a fixed entity with few major design changes that would impact parts as much as other hardware makers. Sure, this year, the iPhone has a larger screen and form factor, which surely affects production lines and parts, but if history holds true, Apple will likely keep this size and shape for a few years and reap the benefits of superior supply chain management and economies of scale.

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

• Tablet market to hit over 377 million units by 2016
• CES 2012: a recap and analysis
• Connected world: the consumer technology revolution

It’s not just the industry shift toward solid-state drives replacing slower, less-efficient hard disk drives that’s driving Violin’s value through the roof, though, it’s also the company’s very strategic set of investors.

The latest investor to get on board is software giant SAP, whose mission-critical ERP and database products are deeply entrenched within large enterprises and could benefit greatly from the price-performance increases Violin’s systems offer over traditional HDD- and-DRAM-based systems. There’s also SAP’s new HANA  in-memory analytic database, which is the focal point of SAP’s big data push. “To the extent that people view HANA as a big data technology,” Basile said, “it fits into our core thrust that we started last year [around big data].”

That “thrust” Basile mentioned is to build the “biggest, fastest, densest” big data systems in the world for technologies such as HANA, Hadoop and NoSQL databases. Violin has worked with various vendors to eliminate the I/O bottleneck in Hadoop that limits throughput into the processor to a few hundred megabytes per second. Using a Violin system, he said, “instead of needing 100 Hadoop servers, you might need 10.”

Violin also counts Juniper Networks among its strategic investors, although flash-memory manufacturer Toshiba might well be the most important. While other enterprise flash-storage vendors must react to consumer demands that regularly change what flash fabricators produce (generally, smaller, less-reliable NAND memory products), Violin gets to plan ahead. Basile said Violin manufactures four generations out because it knows what Toshiba has coming down the pike. Being the company’s only U.S. investment and one of its only strategic tech investments in years, “if you do business with us, you’re also doing business with Toshiba,” Basile said.

Of course, Violin’s chances of a successful IPO aren’t hurt by the story of a Fusion-io, a somewhat competitive company that sells solid-state components that plug directly into servers and serve as high-performance caching layers. Fusion-io has been experiencing major revenue increases and, as of late February, was trading at 160 times earnings.

As Basile explained, everyone involved in the flash space is riding the wave of ever-lower flash memory prices and the entry of mega companies such as EMC, Oracle (e orcl) and HP into the space. “The argument [for flash adoption] becomes not if, only when,” Basile said. “And then, how much?”

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

• Infrastructure Q1: Cloud and big data woo enterprises
• Amazon’s DynamoDB: rattling the cloud market
• Infrastructure Q4: Big data gets bigger and SaaS startups shine

Solid-state drives increase storage I/O performance while cutting power consumption compared to spinning disk and other traditional storage. That means faster launching of applications and faster access to data, but SSD technology has been seen as too expensive for broad use. That is starting to change, however, and even some cloud providers offer at least some SSD options. SSD storage underlies Amazon Web Services’ DynamoDB database, and CloudSigma added SSDs to its public cloud offering last fall.

Although it’s on-premise or hosted, Morphlabs said its new mCloud service will be priced on par with Amazon Web Services, but will bring with it the security of a private cloud to service providers and other companies that require it. Many companies still aren’t ready for the public cloud because of data security and other concerns about shared infrastructure. That’s the crowd Morphlabs is targeting with the new mCloud, which sits upon a Dell converged infrastructure unit featuring the company’s cloud-inspired PowerEdge C servers.

The appeal of a private cloud — or at least a hybrid cloud that lets companies keep some applications and data under their purview — is one reason Amazon partnered with Eucalyptus, which will makes it easier for companies to link Eucalyptus-based private clouds with public AWS cloud services.  Amazon has announced several other services, including a storage gateway and workflow services, all geared to ease interaction between its public cloud and customer data centers.

As with any hot new technology, SSDs started out as a premium option. But as costs come down, expect to see them replace spinning disks more and more both in the cloud and on-premises.

Photo courtesy of Flickr user theaucitron.

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

• Big Data, ARM and Legal Troubles Transformed Infrastructure in Q4
• In Q3, Big Data Meant Big Dollars
• Infrastructure Overview, Q2 2010

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

• Who and what to watch in the new era of the living room
• CES 2012: a recap and analysis
• Connected world: the consumer technology revolution