4 ms·

I agree that storage, which is to be the focus on this article, seems to be a race to the bottom. But when it comes to computing, EC2 and friends are still many

by latch 12y ago

I agree that storage, which is to be the focus on this article, seems to be a race to the bottom. But when it comes to computing, EC2 and friends are still many times less efficient (price/perf) than non-cloud solutions (cheap vps', dedicated or colo). A lot of these value added services sit on top of this infrastructure, so you end up paying quite a bit more than you would have in a non-cloud era. But people say: well, you're outsourcing your devops. This is absolutely false when it comes to the pure compute offering. At best, an EC2 instance is as easy to manage as anything else. It's more true of things like SQS, SES, RDS, DynamoDB, ... though personally, I cringe when a team of developers can't manage their own production postgresql or rabbitmq servers. Still, clearly people are seeing some value in these services. Softlayer, Rackspace and friends dropped the ball here. 15 years of fat profits rendered them sterile. Go back 6 years, sell the performance of bare metal, add quality and affordable managed services (not a $1000/m load balancers), and you might have been able to compete. Case in point: Softlayer's listed prices are still complete bullshit. If you work their sales, you can get servers for 50-60% less even on small orders. But you need to spend 10 days emailing back and forth. Can you even get prices from Rackspace's website? I just went on their site and it made dell.com look useful. IBM should have bought DO or Linode.

MartinCron 12y ago
at best, an EC2 instance is as easy to manage as anything else If you use EC2 in a very basic way, yes. When you start to factor in things like CloudFormation, auto-scaling, elastic load balancing, route 53 DNS, IAM roles, there is a lot of just plain cool shit as part of the AWS ecosystem.
- latch 12y ago
  My point was that there was nothing stopping every other company from doing this. Not in 2006, certainly not in 2012. Some of those companies, then and now (??), are better positioned to truly drive towards zero because of how awful the price/perf is of cloud hosting. Storage aside, it's DO/Linode/OVH/Hetzner, not Amazon, that's driving towards zero. On an aside: Route 53's innovation is that, in huge number of cases, it's dirt cheap. But there have been relatively good and affordable DNS solutions before and since. I am a Route53 fan though, in large part because it's decoupled from AWS. You can use it (and S3) while avoiding everything else from AWS. Auto-scaling is what a sales engineer tells a C[ITE]O in exchange for a signed contract. Scaling comes from design/"architecture"...you can't auto scale your way out of bad design: stateful services, SELECT N+1, n^2 algorithms, LIKE '%spice', poor/unknown cache hit ratios and so on.
  - MartinCron 12y ago
    I personally think that AWS auto-scale groups are poorly named, as they can do more than just the sort of brute force "scaling" that you are rightfully disdainful of. You can, for example, create an auto scale group to ensure a minimum of n healthy instances of an application, so if one falls over, it will be replaced with a pristine one. To me, auto-healing is a lot more interesting than auto-scaling.

by latch 12y ago

MartinCron 12y ago

at best, an EC2 instance is as easy to manage as anything else If you use EC2 in a very basic way, yes. When you start to factor in things like CloudFormation, auto-scaling, elastic load balancing, route 53 DNS, IAM roles, there is a lot of just plain cool shit as part of the AWS ecosystem.

latch 12y ago
My point was that there was nothing stopping every other company from doing this. Not in 2006, certainly not in 2012. Some of those companies, then and now (??), are better positioned to truly drive towards zero because of how awful the price/perf is of cloud hosting. Storage aside, it's DO/Linode/OVH/Hetzner, not Amazon, that's driving towards zero. On an aside: Route 53's innovation is that, in huge number of cases, it's dirt cheap. But there have been relatively good and affordable DNS solutions before and since. I am a Route53 fan though, in large part because it's decoupled from AWS. You can use it (and S3) while avoiding everything else from AWS. Auto-scaling is what a sales engineer tells a C[ITE]O in exchange for a signed contract. Scaling comes from design/"architecture"...you can't auto scale your way out of bad design: stateful services, SELECT N+1, n^2 algorithms, LIKE '%spice', poor/unknown cache hit ratios and so on.
- MartinCron 12y ago
  I personally think that AWS auto-scale groups are poorly named, as they can do more than just the sort of brute force "scaling" that you are rightfully disdainful of. You can, for example, create an auto scale group to ensure a minimum of n healthy instances of an application, so if one falls over, it will be replaced with a pristine one. To me, auto-healing is a lot more interesting than auto-scaling.