Newvem does not provide any information on how it defines the ‘light user’
category, but [...] The company is also somewhat vague about the area under
consideration -- do the figures refer to computing power (EC2), storage
(S3), or the whole AWS package combined?
There is really nothing of substance to see here. On the other hand, the main point being made is probably not wrong. It makes sense for most server owners who only experience light traffic to have 99%+ of their EC2 capacity unused.
For example, I rent a Micro instance for all my stuff and server load is mostly 0.00 with less than 3% of CPU utilization - that's a couple of dynamic websites with about 8-10 HTTP requests per second overall. I'm willing to bet there are a lot of customers out there having an even lower traffic rate and Micro is the smallest server you can rent so of course there will be a lot of unused capacity.
Out of curiosity, how much does that come to per year?
Looking at Amazon's EC2 pricing(http://aws.amazon.com/ec2/pricing/), unless I'm missing something, keeping 7 reserved instances costs about $1000 per year. Even for a small startup, that's a tiny amount of money.
And even if you're using S3 and other services, that's still in the hundreds of dollars per year per service. Again, that won't break the bank, by any stretch of imagination, especially since the cost of maintaining your own equipment would be easily as much, possibly more.
I believe you misunderstood my intentions. I wasn't arguing about whether or not EC2 is too expensive, I was merely trying to guess what kind of reasoning went into the original article.
Also, beside the point, but actually running 7 reserved instances will cost way, way more than $1000 per year.
The article seems to indicate that a lot of people aren't utilizing 100% of their instances and are just wasting money. I'm honestly curious how much this could be.
Obviously it wouldn't make sense for people who only use micro instances, but then again people who only use one micro instance don't use a third party cost management tool.
Yes this is correct. Those numbers are based on data published at http://www.newvem.com/topic/cloud-radar. For reference, light user are those with up to 8 instances, medium users have between 9 and 35 instances, and heavy users above 35. We didn't publish the total number of users in each segment, but I can confirm that that the survey was based on a sample of around 200 customers, as mentioned in GigaOm's article
The reason we chose to segment by number of instances is that we consider this as a good metric for the complexity of one's cloud footprint, and the corresponding effort required to manage this footprint.
Well, of course it's a sales pitch in this context, that's why there is no actual data to back up their claims. But it stands to reason that customers using a lot of instances for high traffic sites can generally budget closer to the actual demand compared to small-timers who probably make up a large portion of AWS customers.
Most of the excess charges I saw during my AWS consulting were due to dev/test environments with fat 64-bit instances that were left running overnight and during the weekends.
Why? It's relatively easy to stand up environments in the cloud, but shutting them off and redeploying takes forever. There is no efficient way to turn complex stacks off and on without a massive time investment from the cloud ops team.
OpDemand solves this with a management toolbar that includes stop/restart functionality. Under the hood we use a combination of EBS volumes and configuration management to allow users to put their cloud environments on ice when not in use. Restarting everything on Monday is a 1-click affair.
OK, this is probably correct, but so what? Keeping a prepaid micro and/or small instance running all of the time is more a matter of convenience than cost effectiveness.
Looks like a study driven by a marketing department.
I think there's an assumption here that idle or stopped servers are bad in themselves. There's definitely a cost associated with spinning up and configuring a new instance, so I don't think this assumption always holds true.
For example, I rent a Micro instance for all my stuff and server load is mostly 0.00 with less than 3% of CPU utilization - that's a couple of dynamic websites with about 8-10 HTTP requests per second overall. I'm willing to bet there are a lot of customers out there having an even lower traffic rate and Micro is the smallest server you can rent so of course there will be a lot of unused capacity.