Hi Steven, Thanks for the insights, very interesting. It is good to know that people with your profile are on this mailing list.
What is your take on persistency and virtual hardware ? The performance difference between say Postgresql running on real hardware vs on ec2+ebs is challenging to say the least. Sven On 24 Jun 2013, at 16:56, Steven Baker <[email protected]> wrote: > I've had quite a lot of experience helping teams who use both cloud-based and > physical-server hosting. > > The promise of the cloud is that you can start/stop servers to meet dynamic > demand. In the simple case, run ten servers during the day in your timezone, > and only one at night to keep the lights on. In a slightly more complicated > case, have some software tools (that you probably have to write yourself) > that spin up and add new servers to the pool when server load reaches a > certain threshold. In the usual case, spin up new servers to do > computationally intensive background workers. > > Since you typically pay for cloud services by the running hour, very careful > management of this can lead to lower costs. In theory. > > In practice, people don't really do this. They spin up some servers, deploy > their stuff. When they need more, they spin up some more, and never really > manage growth or two to manage expenses. In fact, because it's *so* easy to > spin up a new server, folks will often end up with far more than they need > very quickly. > > My experience has been that the average non-profitable startup will spend > between $10,000 and $30,000 per month on their cloud hosting. For hosting > that can be adequately served by two $2500 servers for $250/mo in co-location > fees. I have literally gone to clients, moved them off of cloud provider to > real hardware for significant and nearly immediate gains. One month of their > hosting fees can typically pay for my time (and I'm not cheap), the new > hardware, and a year of co-location fees. > > There are rare cases when this is reasonable. Also, since I'm in Canada, > companies typically can't store their data in the US because of the Patriot > Act, and that makes Rackspace Cloud one of the few options we have. Finally, > you can install VMware's vCloud Director (or similar; they have many > products) on your own hardware and have "the cloud" on your own hardware. > It's amazing. > > That said, there are some good reasons to us cloud-based services: hardware > management, backups, security. > > - Hardware management is the most obvious. When you use a product like > Rackspace Cloud, you don't ever have to think about disks, servers, > locations, network configurations, etc. When you want more power, you click a > button. Someone else manages all of that. When there's a bad disk, you don't > notice, because it's someone else's job to swap out the bad one and make sure > that everything was redundant. > > - Backup and reliability are a really important feature that cloud providers > offer. All will offer redundant disks and the like, but most go further than > that. Most cloud providers have multiple datacentres and can spread out your > servers, or move your data between them quickly. In Canada, we have one cloud > provider who will transparently fail over your stuff from Vancouver to > Toronto and vice versa. > > - Security is probably the most often overlooked point. Most cloud providers > will manage firewalls, and do the "infrastructure bits" of PCI compliance and > related things which significantly reduces the work you have to do yourself. > > In the end, it depends on the business. My experience is that it's foolish to > use cloud providers for anything more than small VPSes without taking a hard > look at the costs. Most businesses will have someone on staff who can manage > the building and deployment of these servers, and well managed servers don't > need babysitters anyhow. Plus, as software developers, we'll automate > everything (they call that Devops now) and make it even easier. > > I have recommended two past clients switch to the cloud. They had a rack full > of aging servers, and were paying a staff of four system administrators ~ > USD$80k/yr to manage those. Moving to the cloud was cheaper for them in both > the short and long term. But it's rare. > > Finally, when your stack is "simple Rails app running on MySQL", which is a > completely boneheaded stack in the first place, it's *easy* to scale and can > be done nearly transparently. Plus, do you really want the Rails kids > touching infrastructure bits? Some things are best left to the experts, and > using a cloud provider makes that easy even if it is more expensive. > > Cheers! > > -Steven > > > > > On 2013-06-21, at 8:29 AM, Stephan Eggermont <[email protected]> wrote: > >> When I see the enthousiast reactions to the cloud hosting tip, >> I often wonder why? What do you see as the advantages, >> and where does it provide a good fit? I can understand the >> 5 euro/month vps for a single instance case. >> >> As soon as I need redundancy, monitoring and scaling, >> I still need to understand so much of the underlying infrastructure >> that the step to DevOps seems awfully small. Neiher Heroku, >> nor AWS make scaling transparent, AFAIK. That is to say, they'll >> start up new instances just fine. That only helps when processing >> is the problem, not internal bandwidth. And solving the wrong >> problem only leads to high costs. >> >> The cost figures I've seen published on the web suggest that most >> larger scale usage would be better of with at least their regular demand >> covered by dedicated servers. I know hosting in the US is much more >> expensive than in Europe. Digitalocean at least seems to give more >> reasonable price/performance. AWS vs Hetzner seems uncomparable >> >> Your thoughts? >> >> Stephan >> >> >> > >
