I have learned a great lesson I wanted to share:
Commodity hardware trumps proprietary hardware.
Last Fall, DataScaler shifted to delivering our software on a cluster of commodity servers. Previously, we had planned to deliver our software on purpose-built custom hardware. The move to commodity servers has been one of the best decisions we've made as a company. Why all the love for commodity hardware?
- Customers love commodity hardware. I was surprised how strongly our customers preferred the commodity servers option. When presented with both options, customers overwhelming chose the commodity hardware, even though our appliances were very elegant, compact, etc. Some customers are HP shops. Others are IBM shops. Some are Dell shops. Their IT teams know how to purchase, configure, rack, and manage their standard servers. Bringing a piece of foreign hardware into their environments is only done grudgingly.
- Competing with HP, IBM, and Dell is dumb. With the major server manufacturers spending billions on server R&D, a small startup cannot hope to keep up. Our (very cool) appliance design packed a lot of compute power into a small space and did it with out melting. The appliance would have given us an 18-24 month lead over commodity hardware in terms of compute density. However, that lead is too short to recoup the costs involved. More importantly, commodity hardware will continue to get better every year, whereas our design would be stagnant. We could have invested many millions each year to update the hardware, but why? The incremental gains would be small and it's nearly impossible to keep up with the giant server makers.
- Commodity hardware is much more capital efficient. By switching to software running on a cluster of commodity servers, we substantially reduced the overall capital needs of the company. Most importantly, we can get to market on less capital, allowing us to raise less money and give up less equity in the early phases.
Luckily for us, there was very little effort involved in porting our software from our purpose-built hardware to commodity servers. All our intellectual property is in our software, and the hardware, though cool, was not core to our business. Shedding our custom hardware has been a great decision, and I only wish we had done it earlier.
As they say, hindsight is 20/20.
