How Google Compute Engine stacks up to Amazon EC2
gigaom.com
gigaom.com
I'm pretty sure that most people who have used any of Google's pay for services knows just how bad that support can be.
What happens if something goes wrong with your billing on EC2? No problem, you call them up and sort it out.
What happens if something goes wrong with your billing on any Google product? You'll be down for days while attempting to get Google to even acknowledge a problem.
Until Google fixes their support issues, anybody using a Google service for anything business critical needs to be very careful in weighing up pros and cons.
Free tier - occasionally get answers. answers are generally vague. questions to forums. Paid tier - always get answers. answers are generally vague. you get a support portal. Top tier paid - always get answers. answers are generally vague. amazon employees will fly out to your site to give vague answers.
(I mean, beside the 'default' ones)
Well, support is relative. If you found Amazon's "horrible", you would find Google's utterly horrible and scary beyond any description.
I think that this comment is spot-on. Google doesn't seem to understand that the infrastructure business is fundamentally a trust business: as elastic and transient as the cloud can feel, when choosing a foundation, you're making decisions that must stand for years. Not only does Google provide infamously Kafka-esque service when they're down or otherwise unavailable, they have a history of capriciously killing services and/or repricing them -- and one need look no further than the user dissatisfaction over GAE to know that GCE will be lucky to survive its infancy. Indeed, GCE is caught in a nasty catch-22: because of Google's track record of killing tertiary services, the market is very much waiting for it to become primary for Google -- but because the market is waiting, it may well never be. This is a self-fulfilling prophesy that is as obvious to cloud decision makers as it appears to be lost on Google.
Or that they wont close it down, like they did with Google Code, Google Wave, Google Reader and such, if they change their minds?
Now, I know that what I wrote would be considered FUD if it was for any other company. But in Google's case, it's not that people say FUD about them. It's that FUD is the very thing they emanate themselves!
Or, in other words, in Google's case, the FUD is real.
After GAE and treatment of "non-essential" products under Larry, I have lost the trust in Google as a reliable business partner, regardless of any technical advantages their offerings might have at the moment.
A small shop like ours can't rely on a partner who can change major aspects of a product without any concern for its customers. And when the trouble comes, being left at the mercy of a random Googler who might come to the rescue if you are sufficiently Internet famous is not a recipee for building a stable business.
Thank you, but no thank you. I'll stick with companies who are relentlessly working on improving their products, driving prices down and care about their customers even if their products might be behind in some areas.
Until Google proves that they care and are in this for a long haul, my business goes somewhere else.
Oh, yeah, this applies to personal usage too - good bye Google Reader, my most used Google product...
I realize that this is mainly just the difference between newer technology and decades of standardization, but until cloud providers reach that point, the prospect of bad/missing support is really frightening.
http://googleappengine.blogspot.com/2013/02/google-cloud-pla...
Sign up here: https://cloud.google.com/support/packages
Basically, it's a way of agreeing with what I think is your point. Support does matter, and Google is taking it very seriously.
Would love to hear your thoughts once you've tried one of those support packages!
It's completely fair to expect customers to solve their own problems, but when the problem is on Google's end, it's beyond outrageous to charge a premium to resolve those.
I'll note a HUGE factor is missing: the word 'cost' isn't even in the article. How much more (or less) are you paying for the apparent increase in performance?
If you want real useful info, go to a industry site not a pop tech chat board like HN.
We’re not quite sure why AWS doesn’t support this, but images on GCE are multi-region (“multi-zone” in their terms)"
Whilst images are not multi-region, you can now copy them[1] across regions with AWS.
[1]http://aws.amazon.com/about-aws/whats-new/2013/03/12/announc...
However, when you factor in EC2 Reserved Instances pricing, the EC2 price drops to $0.028/hour. For a small startup with a couple of instances to start with, you can run your entire EC2 setup for the price of one entry-level GCE instance.
Seems like unless you are going to need the ability to quickly scale the number of instances up and down it makes more sense to put your core services on dedicated servers rather than in the cloud.
Which is almost always the case for startups (needing to scale at will).
How even is a comparison when GCE hardly has any customers sharing the same physical host and network at this stage? It's rather interesting that EC2 read I/O performance was higher despite this consideration.
What will happen when you start sharing a physical host on GCE with many other customers? Will you all still get 157 MB per second disk write throughput simultaneously?
What type of instances were used (and were they EC2's first generation or second generation instances?). Given the fact that they are looking at ephemeral disk performance it's probably a 1st generation instance.
I see no mention of EC2 EBS provisioned IOPs. If performance was a goal then surely having guaranteed provisioned I/O should have been considered? Does GCE even have that feature? Surely IOPS are more important than sequential I/O for most tasks?
When benchmarking EBS I/O were EBS-optimized instances selected?
Why no mention of EC2 cross region AMI copy? Also GCE is currently in the US only. EC2 is on 5 different continents currently.
What are the costs of the instances (using reserved instance pricing for EC2)?
Not correct. They've had EU zones since November:
As the analysis stands, the individual IO stats aren't meaningful. EC2 instances offer a range of I/O throughput in different instances: a t1.micro has 'low' I/O, an m1.small has 'moderate' and an hi1.4xlarge offers 1.1 GB/s write and 2.0 GB/s write. Cost has to figure into a benchmarked analysis here.
And CPU. One of Google's big claims at launch was that CPU time was notably less expensive than on EC2. I'd love to see some data here and clicked the link expecting to find that analysis.