Amazon EC2 – (expensive) servers ondemand
Posted on | August 24, 2006 | 5 Comments
Amazon broke todays biggest news (via TechCrunch) by announcing EC2 (‘Elastic Comute Cloud’) – a hosting service allowing customers to instantly add virtual servers on-demand.
using a locally executed java-tool, system-administrators can configure individual server-images (nicked ‘AMI’, Amazon Machine Image) or customize default-templates. currently there is no definite info on which operating systems will be supported, but I’ld suspect a variety of linux-based systems (though the concept should also work fine with Microsoft-systems). the image is deployed to S3 (Amazon’s Simple Storage Service launched earlier this year) and instantly booted up & ready for action. EC2’s virtual servers equal a physical system equipped with 1.7Ghz Xeon CPU, 1.75GB RAM and 250MB/s bandwidth.
cost structure for running an AMI is a bit complicated: $0.10 per running hour (= $72 per month), $0.15 per GB/month (that’s the usual price of S3) and $0.20 per GB of traffic (S3-price, again).
I compared EC2’s pricing to server4you, one of germany’s most significant low-cost providers. for monthly EUR 40,- (=$51) you get a slightly weaker virtual machine (2Ghz CPU, 0.75GB RAM) including 60GB Harddisk and 3.000GB monthly traffic (additional traffic comes at EUR 0.29 = $0.37) at these metrics, EC2 would cost $72 (CPU) + $9 (storage) + $600 (3.000GB traffic) = $681! phew!
in a nutshell: if Amazon wants to compete with low-cost providers aiming at small/medium-traffic sites (which they probably don’t want to, anyway), they will have to grant large amounts of free-traffic. for high-traffic servers (6TB and beyond), EC2’s pricing already is competitive. there’s no doubt that virtual servers ondemand is an intriguing concept with the potential to shake up IT-infrastructur – for now EC2 is only available to a limited number of Amazon Web Services-customers.
Comments
5 Responses to “Amazon EC2 – (expensive) servers ondemand”
Leave a Reply
July 11th, 2007 @ 5:01 pm
You didn’t factor in the fact that you need servers/processing power too. Bandwidth is not everything. The server-specs you mentioned would break in as few as 10000 simaltenous connections; even if they were using as little as 5KB each (less than 50MB bandwidth).
You need a scalable underlying architecture. That’s what AWS provides you. It is not for hosting small business websites with no signficant computing needs (which I admit probably make up 99.5%+ of all websites).
AWS is for websites which intend to become sort of web-service/product and anticipate a lot of users (and hence computing resources).
July 11th, 2007 @ 6:10 pm
thx for your comment chinmay!
agreed, my comparison wasn’t too fair in that I didn’t consider scalability. (escpecially since server4you is really low-cost). however, at the time of writing, traffic expenses with S3 really seemed like a dealbreaker (haven’t updated myself onthe prices since that). and since I’m running a high-traffic site (http://play.fm an high quality ondemand audio-archive for electronic music) that’s really an issue…
July 11th, 2007 @ 7:17 pm
Thanks, Subnet, for your for your quick response
Even for an on-demand app like play.fm – it seems like bandwidth charges are not deal-breaking expensive. served4u sure does give you a lot of bandwidth but not enough processing power to use it. You’d need multiple served4u servers (and pay separately for acquiring and deploying load-balancers in hardware cost and in technical design expertise cost) to use that bandwidth. Overall, I suspect you’d end up paying a lot more in setting-up the architecture than if you were to use highly-scalable pre-built and proven architecture given by Amazon. Of course, I could be wrong in my cost-estimation of setting up a distributed scaled-out architecture like that of Amazon’s.
Here are some informative case-studies:
Inside MySpace: http://www.baselinemag.com/print_article2/0,1217,a=198615,00.asp
Inside eBay:
http://www.addsimplicity.com/downloads/eBaySDForum2006-11-29.pdf
Google has no such service yet but I think Google’s resources are even much bigger than Amazon and they may be in a position to offer similar Elastic Cloud and S3 like service in future. So we may see some competition and price reduction there in future.
Last but not the least, as far as my knowledge goes (and I ran a web hosting company for a brief period) the whole webhosting industry is based on overselling. Most users don’t actually use even 0.001% of the resource alloted to them. So, Webhosts allot you a lot of bandwidth and resources which they know you are not going to use, and they sell the same resources/bandwidth to multiple clients. Nobody complains because nobody ever uses enough resources to stress out the servers.
Thanks,
Chinmay Arora
July 12th, 2007 @ 1:54 pm
thx a lot for those links, very indepth expertise & impressing numbers going on there! I’ll definitely go through & probably post my opinion, if ‘ve something to add
I totally agree with you: the cost of deploying load-balancing (being it through hardware or software) and server-clustering was totally left aside in my naive/simplified calculation
, I’m aware of it though. what I still wonder: how do Amazon’s EC2-servers scale up? lets say I’ve created one virtual server. when I’m running out of processing power, can I upgrade the existing server in a transparent way? or do I have to setup a second, third etc. virtual server? in the second case, I’ld still have to take care for load-balancing, database-clustering etc. myself, right?
btw., prices for S3-traffic seem to have been lowered since last year, its now 0.18 usd/gbyte…
& thx for the interesting discussion!
July 12th, 2007 @ 3:32 pm
Thanks, Subnet. Yes you are probably right. You’d have to use some sort of load-balancing at Amazon. I am not sure how it works though. I don’t think they give you physical access to data-center. So, how would you go and install hardware load-balancers there – no idea. EC2 is obviously using some sort of load-balancing in their data-center (by virtue of being based on horizontal scale-out architecture). I don’t know if you’ve to implement additional load-balancing for your applications or if Amazon makes their solution available to you.
I believe you’ve to keep adding virtual servers. It would be a scale-out approach rather than vertical scale-up. Similar to that of Google’s, Ebay’s etc.
Traffic between S3 and EC2 is free though. They charge for traffic in/out of Amazon’s environment. So, if you were using database on EC2 and wanted to transfer it out to S3 (persistent storage), then that transfer would be free. EC2 loses all its data the moment you decommission the virtual server. So, you are better off periodically transferring data from EC2 to S3.