Core Infrastructure by Yi-Jian Ngo

Cloud Federation

2008-07-17T05:18:00Z

Chinese mythology is replete with deities that ride clouds up into the heavens. Many startups similarly seeking nirvana on the back of cloud infrastructure have found their ascent rudely interrupted.

It seems unlikely that any single cloud infrastructure provider can be entirely immune to outages. The massive scale of such operations is unprecedented, so statistically catastrophes are to be expected. And there’s the Dark Side of the web, already capable of taking out the internet infrastructure of a small country. By comparison, crippling a cloud provider for fun or profit should be a walk in the park.

While starving startups have few alternatives, businesses are likely to take a dim view of such unscheduled downtime. Of course, they could choose to maintain a duplicate copy of their systems on-premise, but that will really screw up the economic advantages of using cloud infrastructure in the first place.

But what if the duplicate copy lived at a second cloud provider, so if the first cloud provider went belly-up, the systems would automatically failover? Even better, how about an abstraction layer that knits together all cloud providers, so you can just deploy your systems to this “meta cloud” which would automatically distribute it across multiple cloud providers?

Well, we could wait years for cloud interoperability standards to (hopefully) emerge – kudos to Eucalyptus for starting the ball rolling. In the meantime, this feels very much like a wide open opportunity for aspiring cloud startups to step into.

Auto Startups: The Road Ahead

2008-07-09T06:56:00Z

Purveyors of plush mattresses are quick to point out that since you spend 8 hours per day on their product, their eye watering prices are fully justified. Rarely mentioned is the fact that you’re unconscious for most of those 8 hours.

Less well known is that the typical US commuter spends 1.5 hours per day in their car, which is about 3x the average time spent on the internet. Multiplied across 600 million vehicles, that’s a huge chunk of “dead time” that folks are wasting stuck behind the wheel.

In parsing the market opportunities for tech startups building consumer products/services for the automobile, I find it helpful to think of the car simply as a very big consumer mobile device, with many parallels in particular to the mobile phone. The usage scenarios are similar – location based services (navigation, local search, traffic, tracking/SOS), entertainment (audio, video, games) and communications (diagnostics, maintenance alerts). Both even have a similar bugbear with distribution – while mobile startups have the wireless carriers, auto startups have to contend with the auto manufacturers.

Of course, there are important differences between the car and other mobile devices – in particular, the user is always multitasking & has restricted freedom of motion. While it will be foolhardy to speculate on what the next auto-based killer app will be, I believe there are a handful of factors that will increase the probability of an auto startup’s success.

First, user interactions with the service should primarily be auditory – perhaps you could talk to your car and ask it to play a time-shifted radio program. While I’ve come across intriguing systems that can project various images onto the windshield, I personally believe that visually distracting interfaces can be fatal for the driver.

Second, the ability to execute with zero help from auto manufacturers. This does not mean ignoring them – they are a crucial part of the value chain that will need to be engaged at some point – but rather that the business should not pivot entirely on “winning that big distribution deal with GM”. So perhaps making some sort of device sold through retail channels, or maybe software that allows existing devices (e.g. mobile phones) to perform something useful with respect to a car, such as enabling the user to pipe music from a virtual jukebox in the cloud through a car’s audio systems.

Finally, an effective method for in-car advertising that can sustain the business. The reality is that most auto startups’ products will fall in the “vitamins” rather than the “antibiotics” category, and charging a subscription can really crimp adoption. All that excitement around satellite radio seems to have petered out of late, and anecdotal evidence suggest that ONSTAR’s renewal rate is rather dismal.

Lanchester's Laws for Startups

2008-07-02T06:40:00Z

Interesting rule of thumb based on military operational research performed by Frederick Lanchester during World War I:

If you’re trying to enter an existing market where the dominant player has greater than 40% market share, you will need to spend at least 3x the sales/marketing budget of that leader to succeed in a head-on attack.

However, if the largest player has less than 25% market share, all you need is 2x the sales/marketing budget of the leader to succeed – the cost of entry is much lower.

Thus, if you're a startup confronting a formidable incumbent that you lack the financial wherewithal for a frontal assault, consider re-segmenting the market to create a submarket where your product can be unique or substantially different. As an agile but (usually) dirt-poor startup, it’s crucial to pick your battles carefully.

The Art of Raising Capital

2008-06-25T05:32:00Z

Along the road from starting a business to (perhaps) getting to the Next Big IPO, one of the more daunting assignments that an entrepreneur will likely face is the process of raising capital.

The task can be decomposed into qualitative & quantitative elements. The former has many parallels with serious dating. Get a mutual acquaintance to make an introduction. Present yourself well, but be sure to also check out what others think of the investor. Remember that you may be spending many years together, often under high pressure circumstances, so if something doesn’t smell right or makes you uneasy, don’t be shy to walk away.

When it comes to the terms & numbers, it’s important to educate yourself on the basics. Know your way through a term sheet. Understand the common financing options you have. Talk to others who’ve been through this before. So when rubber hits the road, you have a mental framework ready to guide the negotiations.

I believe that a helpful perspective is to start with the total capital you expect to need at multiple points over the next few years to get your company to where you want it to be, and then back into an appropriate valuation range for today. A useful simulator for visualizing this analysis can be found here.

Microchunk the Operating System

2008-06-18T05:02:00Z

Given all the cacophony around JeOS, and the vigorous (and probably correct) assertions that virtual appliances are the future model for application delivery and deployment, you could be excused for tolling the death knell for server operating systems.

After all, should you desire to deploy an application in this alleged future, you would just download the appropriate virtual appliance, plunk it on your favorite hypervisor and viola! Who needs an operating system (described memorably by FastScale as a “bloated 2GB DLL”) that brings with it a nightmare of configuration mismatches, incomprehensible incompatibilities and other sundry pain points?

The reality, as is often the case, is a little more nuanced. The operating system hasn’t vanished – it’s simply morphed into a more svelte entity that’s comes bundled within each virtual appliance and only contains the components that the application actually needs.

So what’s changed is not the existence of an operating system, but rather how it is packaged & distributed. The era of the one-size-fits-all operating system pushed out through hardware OEMs may be drawing to a close. What we could see instead is a buffet of operating system components that application vendors will selectively mix-n-match and then distribute as part of their virtual appliance.

This transition from a hardware OEM to a software OEM licensing & distribution model will likely be heart wrenching for operating system vendors. Then again, if there are any vanishings in the coming years, it’ll probably be of those operating systems that failed to adapt.

The Promise of Serious Games

2008-06-11T05:15:00Z

Won’t it be nice if you could come in for work and just play computer games all day? Given the tantalizing productivity gains that could be reaped if people applied the same zeal to work as they do to games, it’s no wonder that much energy has been expended on bridging the gap.

One approach has been to bring elements of work into the gaming environment. Particularly promising have been the various training simulators for the military & healthcare verticals, some of which are rather effective and quite literally save lives.

More disappointing have been the multitude of educational games that were supposed to revolutionize the classroom. Their main problem is a gaming environment that pales in comparison to what most youths today expect, which can easily cost tens of millions of dollars to build.

Somewhat intriguing has been the idea of using games for crowdsourcing. For example, players in a virtual world often choose a profession such as a doctor or a miner. In order to raise their level, they could be asked to acquire a slice of real world domain knowledge, such as how to identify a cancer cell or a good place to drill for oil, and have actual data fed to them. The results could provide valuable information to real world physicians and geologists.

The other approach has been to bring elements of the gaming environment into the workplace. Here I believe the greatest potential lies in immersive & intuitive user interfaces. Imagine the time saved if I could navigate my desktop, an ERP system, or for that matter the management console of a datacenter with the same ease that I can plunge through a virtual game environment.

Also interesting has been the idea of creating some kind of points system that allows people to store/redeem/trade credit for tasks they perform – an analog to Dragon Kill Points which are used to distribute the loot from a mission among team members in many role playing games.

Seriosity has already built an email prioritization application based on this concept. Another use could be enabling people within a company to build a reputation for domain expertise, hence improving the accuracy of enterprise people search tools. Or perhaps it could form a component of an internal prediction market that augments existing corporate forecasting methods.

Why Books are Better than Blogs

2008-06-04T06:03:00Z

Even as the size of the blogosphere continues to burgeon, I’m finding it increasingly hard to find high quality blog content. The signal-to-noise ratio has decreased, but the filtering technologies just haven’t caught up. Today, if you want to find original, insightful content on most topics, you’ll probably do a lot better perusing the shelves at your friendly neighborhood library than trawling through the oceans of endless blogs.

There are several reasons for this. People tend put a lot more thought and effort into a book than they would into a blog post. And while it’s technically possible today to self-publish, for the most part books still need to survive the scrutiny of publishing houses to get meaningful publicity & distribution, which puts a floor on the quality of what makes it into your public library.

But the most significant problem for blogs, I believe, is the lack of any ranking system that actually works. In comparison, it’s much easier to separate the wheat from the chaff from a mountain of books. A simple & effective yardstick is how well a book sells – when people have to open up their wallets, they’re making a tangible vote of confidence in what they’re expecting to get in return.

One way to augment that yardstick is by also taking into account when the book was published. Any book that came to press decades ago but is still selling like hotcakes probably has something pretty compelling within its covers.

Today, there are at least a couple of ways that blogs are ranked. The first is by harnessing social energy, but that has been shown to be vulnerable to manipulation. The second is algorithmic approaches which try (and mostly fail) to return both relevant & quality content. One of the better tools I’m stumbled across is filtrbox, but there’s still considerable room for improvement.

One way or another, I think there’s a lot of value in building a service that can, for a particular topic, bubble up the best of the blogosphere. I certainly would pay for it. And I think lots of other people & businesses would too.

In Search of an Online Eraser

2008-05-28T06:58:00Z

It’s been nearly 10 years since Scott McNealy famously admonished a bunch of reporters: “You have zero privacy. Get over it!”

Nowhere has this statement been driven home harder for me than the few minutes I spent on pipl and ZabaSearch, free websites that search & aggregate data about individuals. Suffice to say that they could be quite disruptive to certain segments of the Private Investigator industry.

While most people are still blissfully unaware of the existence of such services, it’s only a matter of time before they bubble up into the mainstream, and there’s going to be a lot of nervous and unhappy folks out there.

I think there’s a huge market opportunity for a service that can discover and eliminate information about individuals on the internet. Execution could be tricky though – this feels like something that’s going to get entangled with regulatory/legal policies fairly quickly.

Social Networking in China

2008-05-21T04:09:00Z

Many American parents cringe when they learn about their offsprings’ behavior on the web. Salacious messaging, saucy photo sharing, revealing of intimate personal details – oh my! But the results of an IAC/JWT survey of 18-25 year olds in urban China and US suggest that American youths are relative church mice:

% agree	China	US
Interactivity helps create intimacy, even at a distance	82%	36%
It’s perfectly possible to have real relationships purely online with no face-to-face contact	63%	21%
The Internet helps me make friends	77%	30%
The Internet broadens my sex life	32%	11%

I believe that these results have several implications. First, China will lead the world in the sales of virtual consumer goods. If people care so much about their online persona, you can bet that they won’t be holding back when accessorizing their digital alter-egos with everything from designer shoes to novelty genitalia. These “frivolous” items, rather than advertising, will likely form the bedrock of revenues for Chinese social networking sites in the near term.

Next, certain categories of relationship-based consumer web services that flopped in the US could flourish in China. One example is online dating, which has a spotty record at best in the US, while in China the dating site 51.com ranks among the top 10 online properties by visits. Another is demand aggregation, where large groups of people are rallied online to visit a store at the same time to demand a lower price for a specific item, sort of a physical variation of defunct sites like MobShop and Mercata.

Finally, there’s the intersection of all this social energy with the heavy usage of mobile phones. I believe that this will lead to many examples of innovation and invention that could push China ahead of the US in the mobile social networking domain.

Quantifying Music

2008-05-13T06:01:00Z

Over the past decade, data driven techniques have been steadily gaining ground over more intuitive approaches. Marriott’s Courtyard brand was born from the results of a conjoint analysis, not a light bulb going off at an executive retreat. And the crisp metrics generated by online advertising campaigns are placing intense pressure on legacy “I waste half of my budget, but don’t know which half” advertising models.

It's particularly interesting when quantitative methods are applied to former bastions of intuition. Moneyball is a memorable account of how a professional baseball team used statistical techniques rather than talent scouts to identify undervalued players. And Epagogix has sophisticated algorithms that can allegedly predict a movie’s box office takings based on the script alone.

It doesn't take an education in music to observe the similarity between musical and mathematical notation. Just as algebraic chess notation is used to describe the moves in a game of chess, I wonder if a piece of music can be expressed in some standard schema that makes it amenable to quantitative analysis. That will open the door for analyzing the entire musical corpus, and perhaps distilling the essence of what it is that makes a piece of music “good”.

The obvious near-term application of data driven musical analytics is to predict whether a new song is going to be a hit – companies such as Platinum Blue and Polyphonic HMI already have some traction. I wonder when we will have the ability to build the musical equivalent of IBM’s Deep Blue–a system which can write original music just as well as the world’s leading composers, and the impact that will have on the world’s music ecosystem.

Managing Startup Founders

2008-05-07T06:03:00Z

I recently had the opportunity to have dinner with a small group of startup CEOs. Unusually, none of them had actually founded a startup before – they were all professional managers brought in to take a startup to the next level.

We discussed the lessons they had learnt over the course of their startup careers (most had led more than one startup), and one thing that kept coming up was how best to manage their relationship with the startup’s founders.

They agreed that the right approach was being firm with the founders that they were in charge, and to quit quickly otherwise. This did not mean pushing the founders aside, rather that the founders should stick to the division of responsibilities that were agreed upon and not engage in non-productive interference, particularly in areas they had little expertise in. Trying to politely ignore or manage around such antics simply prolonged the inevitable, and could seriously hurt the startup in the process.

Metering Cloud Computing

2008-04-30T06:18:00Z

One of my favorite quotes from the book Fooled by Randomness is

Wittgenstein's Rule: Unless you have confidence in the ruler's reliability, if you use a ruler to measure a table, you may as well be using the table to measure the ruler.

I believe that we are in the very early phases of the migration of enterprise IT infrastructure into the cloud. Once it’s there, though, I expect that there will be a need to measure its consumption in a uniform way, regardless of whether it’s desktops, servers or data centers that are being purveyed. Simplistic metrics like time or compute cycles are unlikely to survive contact with cloud infrastructure that can conjure up customized stacks in real time from globally-distributed & heterogeneous resource pools.

Failing to come up with an appropriate yardstick could lead to hairy billing issues, savvy customers tinkering with clever arbitrage schemes and potentially the inability of cloud service providers to effectively predict how much to charge in order to cover their costs.

One candidate for such a universal metric is the Computing Resource Unit or “CRU”, developed by SatoriTech, a virtualization analytics startup. CRU is a quantitative measure of compute capacity calculated via an algorithmic approach that normalizes a wide range of input variables.

Unlike metrics that are simply a thin layer of code wrapped around a set of arbitrary “consultant-generated” frameworks, the patent-pending technology behind CRU is the product of multiple years of research and is currently deployed in some of the largest commercial & government institutions in Sweden, where it is used to measure & benchmark the efficiency of internal IT infrastructure.

I wonder what the impact of such a universal metric will be. Some would argue that it will commoditize IT and destroy value. But it's also plausible that the economies of scale that cloud service providers achieve would merely shift value in their favor and that of their end customers.

Startups & Friction Free Cloud Services

2008-04-21T22:03:00Z

Golf is an activity that requires a tremendous amount of overhead. Aside from actually learning how to swing, aspiring players also have to set aside continuous time blocks of 4 hours, gain membership at a suitable course, acquire an expensive bouquet of clubs & assorted paraphernalia and plow through an arcane set of rules.

Soccer, on the other hand, can be enjoyed with just a ball and some open space. It’s hardly surprising that there are a lot more soccer players than golf players in the world.

One of the things that I look for in a startup is a product or service that’s really easy for customers to adopt. Let’s just say that I’m allergic to anything that’s complex enough to need direct sales forces, armies of consultants and/or meddling with large swathes of the installed IT infrastructure base.

While the delivery of enterprise applications from the cloud has gone a long way to reduce the friction of adoption, the use of cloud infrastructure services themselves to power an enterprise is still pretty hard to do. Nick Mehta from Trinity Ventures has some interesting musings here, but the bottom line is that if you’re the IT manager of a mining company in Wyoming, you’re going to find the task of adding cloud services to your infrastructure about as pleasant as a root canel.

I believe there’s a huge opportunity for startups such as 3tera to make using cloud infrastructure services more like playing soccer instead of golf. A good start will be a set of online tools that have the look and feel of existing management consoles. And the next step could be to make it really easy for the enterprise that needs just that extra bit of compute power/storage/etc. during peak times to “burst” into the cloud.

Can Security Startups Succeed?

2008-04-15T20:47:00Z

These days, you hear a lot more about “consolidation” and “maturity” in the security industry than about innovation. VC investments in security startups have halved over the last 2 years. And a recent survey by Goldman Sachs indicates that the growth of IT security budgets will decline from the double-digits to just 8% in 2008 and 6.5% in 2009

So it would seem that doing a security startup today is somewhat akin to train surfing. I beg to differ.

First, the fundamental drivers of security threats are accelerating. Gary McGraw’s “Trinity of Trouble” describes these as connectivity (growth of the internet & web services), extensibility (widespread use of plug-ins, mashups) and complexity (exponential increase in the length of code bases).

It is unsustainable for the severity of threats to increase at a rate greater than the resources available to mitigate them – that trajectory implies that swathes of IT infrastructure will gradually be rendered unusable. I believe there is the opportunity for innovators to plug the gap with a different approach, possibly spawning the VCs’ holy grail – the next $1B+ security category. While only time would tell the details of its composition, I expect that aspects of automation and semantics will be involved.

Second, security is a perpetual game of cat-and-mouse, implying a continuous need to build new defenses to adapt to the shifting threat environment. So while things like anti-malware and anti-spam are indeed “mature”, the tectonics of technological innovation are exposing new surface areas of attack, from virtualization to rich internet applications to mobile phones.

While many enterprises tend to be more concerned about mature attack vectors than emerging ones, it usually takes just one CNN Moment for corporate wallets to rapidly spring open. And at that point, startups that have developed rock-solid defenses for those new attacks will find themselves well placed.

Organizational Structure for Startups

2008-04-09T03:54:00Z

Had a lively group discussion about the best way to structure a startup’s org chart. Went to and fro on the pros and cons of various models, until a veteran entrepreneur hit the nail on the head – you know that you need to reorganize your startup if your staff are spending most of their time in internal meetings.