Gorillas, spider monkeys and startups: Hitting the scale wall

While at VMworld 2012 last week, I couldn’t help noticing the large number of startup companies, row after row of them. So many small tech companies, most of which I had never heard of before, all struggling to carve a path in an industry crowded with 800-pound gorillas.

We all know the gorillas, the companies that dominate by their heft alone: Cisco, EMC, Oracle, IBM and so on. It’s a top heavy industry. But even so, there are many little spider monkeys trying to grow up to become gorillas or at least to be adopted by a gorilla family to live happily ever after. I’ve clearly taken this metaphor far enough for now!


Credit: Lostonpurpose (under CC licence)

The startups got me thinking about which will succeed and why. Or, looking at it the other way, why will so many fail? There’s no one reason.  As I thought about it, two areas struck me as a problem for startups: scale and pricing. Let’s talk about scale first.

I use “scale” in a specific software sense, namely the ability for a software product to deal with very large numbers of nodes (a “node” is whatever that software is touching: servers, phones, switch ports).  Scale is essential if you are going to sell into large organizations, the kind that will invest lots of money for your product and keep your startup alive for at least a while longer.

I lived the scale problem myself. It was back in the go-go late 1990s during the first wave of internet access. I was working at Ascend Communications, later acquired by Lucent Technologies in one of the biggest acquisitions in tech history.  Ascend was wildly successful and exploded from $16 million in revenue in 1993 to over $1 billion in 1997. Not bad.

Ascend dominated internet dial-up. For those too young to remember the world before high-speed internet, you used to connect using a standard telephone line and a modem. You’d dial a number, your computer made spaceman sounds (buzzz! beep! boop!) and then you were connected to a very slow version of what we used to call the World Wide Web.  On the far end of that connection were, at first, banks and banks of individual modems, nightmares to manage. Ascend put a bunch of modems into a single box. ISPs the world over loved it and they sold like hotcakes. (It should be noted the idea came from one of their customers – it pays to listen!)

Now, enter scale. With dial-up users growing rapidly, ISPs wanted more and more modems in each box. This was one level of scale which culminated in the MAX TNT platform that jammed 720 connections into one box.  I came in on the network management side, trying to monitor those connections. In effect, each MAX TNT had 720 nodes in it, each one kicking out data about connection time, speed, etc.

Managing 720 nodes is not a problem. But ISPs didn’t have one TNT. Some had thousands. Giant dial-up providers – companies like AOL, UUNet, PSINet – deployed rooms full of TNTs, with ports numbering in the hundreds of thousands, handling millions of connections a day. Each connection had to be monitored.

The monitoring software we used simply wasn’t up to the task at the larger ISPs. Even though we ran it across multiple giant Sun servers, it couldn’t keep up with the demands made on it. Features were easy to create, and we had plenty of cool features. But they were crushed under the load.

So what does this have to do with data protection? Everything! The thing about data protection is that it touches all the nodes in the data center. You don’t back up only some of your servers (I hope!). You back them all up. Your node count can easily reach thousands. And unlike monitoring  dial-up ports – a fairly simple task before scale kills you – running backups is a much heavier workload and generates lots of data you need to track.

And this is where startups hit a wall.

One way that startup monkeys stand out in the crowd of gorillas is by innovating. In the backup world, this was very evident when protecting virtual machines. The backup gorillas (you know their names) were slow to respond to virtualization and as a result a bunch of spider monkeys showed up with cool new features the gorillas didn’t have.  Many of them were successful and sold lots of software, but they focused on features and not on scale. And in software, features are easy. Scale isn’t.

Many of the spider monkeys are now hitting the scale wall. The dilemma is that scale has to be built into your architecture from the start, from the first day you write a line of code. If you try to build scale later, you fail. Then you have to re-architect, and that means feature stasis, buggy new releases and annoyed customers.

Meanwhile, the gorillas may be slow but they sure aren’t stupid. They watch the spider monkeys and copy their ideas. And guess what? The gorillas figured out the scale problem a long time ago (that’s why they are now gorillas).  They may not have first mover status, but after a few years they close the feature gap with their already very scalable products.

While this was happening, end users were growing their virtual machine numbers, so they grew in scale too. The monkeys kept giving them new features, but not new architectures. Hello scale wall.

The industry is seeing the shift now with larger organizations dropping their virtualization point-products – yes, sometimes sacrificing features – to roll their virtual machines up under the scalable umbrellas of their gorilla software (a rough metaphor, I admit, but I like the idea of gorillas with umbrellas!).

The lesson: if you want to create software that works in high-node environments (backup, network management, security), you better think long and hard about “how will this scale to a million nodes” even if your first customer has a hundred nodes. Remember the inevitable trajectory: feature, feature, feature – bang! – scale wall.

Next time, we’ll consider the importance of pricing and how it drives innovation. Yes, I said pricing drives innovation. Stay tuned. 

Copyright © 2012 IDG Communications, Inc.

7 inconvenient truths about the hybrid work trend
Shop Tech Products at Amazon