3 ms·

There's also cost. At certain scales (such as Grooveshark's scale) it can be significantly cheaper to do infrastructure in-house. Try negotiating bandwidth pric

by wanderr 16y ago

There's also cost. At certain scales (such as Grooveshark's scale) it can be significantly cheaper to do infrastructure in-house. Try negotiating bandwidth prices with Amazon.

edge17 16y ago
Actually, I talked with them at openworld a while back, and i got the impression that the prices were negotiable if you're a big enough customer. I may have totally misunderstood them though, but I also talked with them back when they were significantly smaller (though growing rapidly).
lsc 16y ago
the big advantage the cloud brings is speed of spinning up new servers, for which it's nearly unbeatable. If you need 1000 servers tomorrow and want to get rid of them on Friday, there really is no other option. However, if you can forgo that speed, if it's okay to take some time to get the servers up and if you plan on keeping them online for a few months, owning starts to make sense around the time you need 16-32GiB ram, which really isn't very much.
- TeHCrAzY 16y ago
  Not only that they can be up today and down tomorrow, but that you only pay for the time you are using them, which is (relatively) insanely cheap: www.google.com/search?q=0.34+dollars+per+hour++4+days++1000
  - lsc 16y ago
    yeah, if you only need a few hours a day, "the cloud" is the place to be. But, most workloads are on 50% or more of the time, (see the popularity of amazon's reserved instances) and you save a lot more than 50% owning your own hardware, unless you are paying ridiculous fees for your hardware guy.
- wanderr 16y ago
  Agreed, however the article is about adoption of cloud services. Hopping on temporarily for extra capacity doesn't seem like the kind of adoption they are trying to encourage. The other problem is that if you run a mix of in house and out of house services, the latency between the two sets of servers may be too high to be of much use. Say for example your DB server is on your network, and you need more capacity on your web servers. You could fire up a few extra EC2 instances and set up a VPN link from them to your DB server, but most likely the latencies are going to be too high to be useful. So cloud services aren't necessarily all that useful for temporary capacity either. To some extent it's all or nothing, at least for web services.
- TomOfTTB 16y ago
  You're assuming either/or. I have a mostly in-house IT infrastructure but I use EC2 not only for projects that I need to get up and going quickly but also for traffic spikes. Using a dynamic DNS like Nettica allows you to blend quite easily.
  - lsc 16y ago
    I think that a mixed setup like you describe can allow you to get the best of both worlds, if all your servers don't need to be in the same place for latency reasons. "the cloud" is also awesome for backups. you can have a full backup cold spare system, test it once a week, and pay almost nothing for it. (this is what those 'micro' instances are for. you have a micro instance updating your DB on an EBS backend. when it comes time to fail over, shut down the micro instance and bring up the ebs mount on something beefy.) my point was just that most systems are on most of the time, and for those systems, owning saves you a lot of money over using "the cloud"

by wanderr 16y ago

There's also cost. At certain scales (such as Grooveshark's scale) it can be significantly cheaper to do infrastructure in-house. Try negotiating bandwidth prices with Amazon.

edge17 16y ago

Actually, I talked with them at openworld a while back, and i got the impression that the prices were negotiable if you're a big enough customer. I may have totally misunderstood them though, but I also talked with them back when they were significantly smaller (though growing rapidly).

lsc 16y ago

the big advantage the cloud brings is speed of spinning up new servers, for which it's nearly unbeatable. If you need 1000 servers tomorrow and want to get rid of them on Friday, there really is no other option. However, if you can forgo that speed, if it's okay to take some time to get the servers up and if you plan on keeping them online for a few months, owning starts to make sense around the time you need 16-32GiB ram, which really isn't very much.

TeHCrAzY 16y ago
Not only that they can be up today and down tomorrow, but that you only pay for the time you are using them, which is (relatively) insanely cheap: www.google.com/search?q=0.34+dollars+per+hour++4+days++1000
- lsc 16y ago
  yeah, if you only need a few hours a day, "the cloud" is the place to be. But, most workloads are on 50% or more of the time, (see the popularity of amazon's reserved instances) and you save a lot more than 50% owning your own hardware, unless you are paying ridiculous fees for your hardware guy.
wanderr 16y ago
Agreed, however the article is about adoption of cloud services. Hopping on temporarily for extra capacity doesn't seem like the kind of adoption they are trying to encourage. The other problem is that if you run a mix of in house and out of house services, the latency between the two sets of servers may be too high to be of much use. Say for example your DB server is on your network, and you need more capacity on your web servers. You could fire up a few extra EC2 instances and set up a VPN link from them to your DB server, but most likely the latencies are going to be too high to be useful. So cloud services aren't necessarily all that useful for temporary capacity either. To some extent it's all or nothing, at least for web services.
TomOfTTB 16y ago
You're assuming either/or. I have a mostly in-house IT infrastructure but I use EC2 not only for projects that I need to get up and going quickly but also for traffic spikes. Using a dynamic DNS like Nettica allows you to blend quite easily.
- lsc 16y ago
  I think that a mixed setup like you describe can allow you to get the best of both worlds, if all your servers don't need to be in the same place for latency reasons. "the cloud" is also awesome for backups. you can have a full backup cold spare system, test it once a week, and pay almost nothing for it. (this is what those 'micro' instances are for. you have a micro instance updating your DB on an EBS backend. when it comes time to fail over, shut down the micro instance and bring up the ebs mount on something beefy.) my point was just that most systems are on most of the time, and for those systems, owning saves you a lot of money over using "the cloud"