I've much enjoyed your illustration throughout the group matches. Interestingly, I've shared it with others and they all fell silent. But once I explained it to them they liked it a lot and shared it with others. Who also fell silent. For many people it seems it's too much information/impenetrable. For me it was a puzzle to solve (as I wasn't current on current round-robin protocol). Really good work. Keep it up and see you in 4 years!
Sorry for just barfing this out. I'm most curious about whether distinctions are made between back end or "infrastructure" code and "application" code. What type of org do you manage or work in and what are/are there policies?
Paul, it's tough to do preliminary research on AWS and summarize it so concisely. Best woud be to start/continue using sets of services and the nuances and "gotchas" you noted will become clearer. Particularly EC2. For example, here are some points you noted followed by --> some clarifications:
* Lots of different sizes for cpus. --> Virtual servers come in many configurations. Choose based on the combination of CPU, RAM and disk I/O required.
* Only scales down to 40 dollars a month. --> t1.micro instances are $15/month. m1.small instances are $40/month or as low as $25/month with reserved instances (amortized over 1 year) or even less than $25/month with spot instances.
* If you reboot it, the data will disappear. --> No. Data is not lost on reboot. Most instances have two types of storage: EBS and Instance Store. The Instance Store data is lost when instances are powered off. EBS and Instance Store data survives across reboots. Understanding the difference between a rebot and a power-off is important.
* Firewall is turned on by default so you can't connect to it. Have to turn that off. --> Clearer would be to say firewall ports need to be opened to support the services you need (such as SSH/RDP, HTTP, etc.).
* The name of it changes on reboot. No. The DNS name, external IP and internal AWS IP are preserved across reboots. They are lost when an instance is powered off. (See above.)
* Use the IPs which amazon gives away for free. --> AWS limits you to 5 Elastic IP addresses (for the most part). You can (for the most part) avoid using Elastic IPs by using the public DNS of the instances, often in conjunction with your own ___domain DNS. For example, a database server might resolve like so:
db1.example.com (your ___domain)
|-> ec2-12-34-56-789.compute-1.amazonaws.com (public AWS DNS)
|-> 12.34.56.789 (public IP if you are outside of AWS)
|-> 10.11.12.13 (internal AWS IP if you are inside AWS)
Enjoy the learning experience. It will take some time for it to all sink in.
Oh, I've been using EC2 + S3 for five years (2007). I was looking in to the newer stuff (beanstalk, dynamodb).
Please add your feedback to the details pads so other people can use it in the future. That is why I posted it on hackpad and not my blog.
EDIT: thanks for the feedback.
Sorry if I wasn't clear of the purpose. I don't love the Amazon documentation and wish people would use Hackpad instead because it would be more up to date.
This is my effort to bootstrap that step by providing and overview and details rather than googling old blog entries ever time I want to setup EC2 with ubuntu.
Do you have any specific criticisms of our documentation that I can pass along to our team? Do you want more, less, or something different? Feel free to post or contact me via the email address in my profile.
When Amazon notes "SSD," from a client's perspective, it is only marketing. The storage media matter when you are managing your own hardware. The storage media do not matter in SaaS. For example, the storage media could be floppys and you would get satisfactory performance if there was a memory cache. Similarly you could get poor performance with SSD media if the networking layer(s) were slow. Similarly if the storage media were fault-likely CD's that wouldn't matter either, to us, because of the data replication performed in "the service." What matters in this case is the reported and actual latency.
The question was deliberately broad and honestly inquisitive. But born of the narrow observation that while there are standardized, published formats that separate style from content (HTML/CSS) and similarly, say, LaTeX, much human "work" is put into formats where the style /is/ the content. Say, a pretty resume in Word or a PowerPoint presentation.
We have conversations where we say things in passing. Maybe we don't care to preserve those words. But when we bother to document things presumably they are worth something. So I wondered how much "worth" was being lost or later mis-read.
Perhaps some or much documentation is simply a means to an end and its longevity is immaterial.
