Amazon Beefs Up Its Cloud Ahead of MS Announcement 89
Amazon has announced several major improvements to its EC2 service for cloud computing. The service is now in production (no longer beta); it offers a service-level agreement; and Windows and SQL Server are available in beta form. ZDNet points out that all this news is intended to take some wind out of Microsoft's sails as MS is expected to introduce its own cloud services next week at its Professional Developers Conference.
Re:What is it? (Score:5, Interesting)
Replace "cloud" with "mainframe" and take 40 years off your age, and then you pretty much have it, as is my understanding.
Re:What is it? (Score:3, Interesting)
In this context "cloud computing" is somebody else managing virtual servers for you, and providing an API to add and remove servers from the ones you have running.
Cloud Computing is Creepy (Score:2, Interesting)
Re:According to the TOS: No Spamming (Score:3, Interesting)
> Subject to the terms and conditions of this Agreement, you may generally publicize your
> use of the Services; however, you may not issue any press release with respect to the
> Services or this Agreement without our prior written consent.
In other words, there will be no negative reviews published.
Watch the license... (Score:3, Interesting)
Amazon has said very little clearly about who is paying for the Windows license in the VM, and what situations you can replicate a Windows VM in.
What they wish you could do and what Microsoft allows you to do (given the need to change SIDs and machine names, and the fact that a VLK can't be used in that scenario) means Amazon is punting license compliance onto the end user who will likely not be able to do what they wish they could do.
Re:What is it? Scaling (Score:4, Interesting)
Add to that the cost.. In our cost comparisons we found EC2 costing the same or more than managed dedicated servers with tier 1 providers.
BAM! That's what turned me off of Amazon as well. Anything they can do, I can do cheaper elsewhere with "conventional" servers. There are literally hundreds if not thousands of hosting companies just dying to lease you a $49/mo dedicated server that runs circles around any EC2 VPS, and most of them have at least 500gb of traffic included in the base price.
For ~$150 I have 10mbit unmetered, on a dual-core Xeon. Actually I have several, with reverse proxies and what-have-you, just like the Amazon cats do when they want to scale. The big differences are: I have static IPs, and my costs are lower. I am at risk of hardware failures, but then again I can afford an extra box or two for redundancy/backups.
I could see EC2 being worthwhile for small or short-lived jobs, but the moment you start talking about multiple instances and pound/squid nodes, you might as well move to a dedicated box.
Re:What is it? Scaling (Score:3, Interesting)
"I have static IPs"
Amazon has offered "Elastic IPs" for quite some time. Once you request an IP address it is yours until you release. It doesn't have to be assigned a particular instance.
The key difference is that you pay for an Amazon instance only as long as it is turned on. If you use dynamic or scheduled scaling (if you have predictable traffic patterns) you can scale up or down your servers as you need. You only pay for the time they are turned on. Obviously it completely depends on the type of application you are running as to whether or not this is an advantage.
Re:What is it? Scaling (Score:3, Interesting)
Your definition of "cheap" amuses me.
Thing is, your dying "$49/month" hosting companies tend to suddenly become *very* expensive when and if you need to scale.
In reality the common route is indeed to start out on such a discounter, until it assplodes. And then rent a rack somewhere, fill it with own hardware and move as fast as you can.
Compared to *this* route (which many startups can sing a song about) the amazon prices don't look so hefty anymore.
And frankly, it is definately that part of the scalability story that you should watch out for.
What does it matter whether you pay $49/month or ~$90/month (amazon) during the early days?
I don't think those $600 bucks that you may save in the first year will ease you much when you get to spent 5 digits on hardware to handle load spikes that, on amazon, would only cost you a few hundred bucks each...
Re:What is it? Scaling (Score:3, Interesting)
That's funny, I've been scaling for years, and my host has been around since the last 90's.
I don't deal in anything that has such dramatic spikes as to discredit my system. I'm perfectly comfortable with the fact that my machines are 80% idle during off-peak times. I'm far more concerned with issues like network congestion and Apache hogging because my traffic is bandwidth-bound, not CPU bound.
Just because all the kids today are writing hungry RoR apps, doesn't mean I'm bound by such ridiculous bottlenecks. Just one of my boxes can serve roughly 100 requests per second, which is why I have to use squid frontends, else the clients end up tying all my Apache threads waiting to finish.
Designing my app to be easily distributed means I can scale Google-style, by adding cheap servers as needed. I also spread out my servers geographically around the globe, with plentiful bandwidth at each location. With EC2, you're sharing the pipe with everyone else. You can't get dedicated bandwidth, and from what I've seen the capacity just isn't there (yet?).
Yes, I'm talking about serving pr0n. Next question.
Re:What is it? Scaling (Score:3, Interesting)
I don't see why you think that ec2 should be discredited as a whole based on a use-case that they're not even remotely targeting?
Ec2 is for application hosting or number crunching. Static content (like porn) is the job for a CDN. Ec2 is not a CDN.
And honestly, claiming you are bandwidth bound, using apache and serving a miserable 100 hits/sec per node all in the same paragraph does not help your credibility much.
You're either doing it really, really wrong (cf. epic fail) or you're just trying to sound important without having ever really touched a large scale system.
For reference, an async-io server like nginx, lighty, zeus or similar will easily saturate a Gbit uplink on moderate hardware, either with small files (then it will push upwards of 1000 reqs/sec) or with large files (then your "100 reqs/sec"-statement makes even less sense). Sorry, but thanks for playing.