Amazon EC2 hosting,

[Maxim Veksler]

On Sun, Oct 10, 2010 at 5:04 PM, Orna Agmon Ben-Yehuda
<ladypine at gmail.com>wrote:

>
>
> 2010/10/10 Ori Idan <ori at helicontech.co.il>
>
>>
>> 2010/10/10 Tom Rosenfeld <trosenfeld at gmail.com>
>>
>> Hi,
>>> I just came across this thread from back in Aug about Amazon's cloud.
>>>
>>> I'd like to add that I have been a satisfied customer of Amazon for over
>>> a year, using their services for both consulting at at my current job where
>>> we use it to run our SaaS offering. The capabilities keep improving and the
>>> prices keep coming down. Their lowest end server is now just 2 cents an
>>> hour!
>>>
>>> There are some issues with the IO, but it is certainly adequate for all
>>> but high performance needs. We use 8 way stripped disks and get about 100
>>> MBp/s sequential reads.
>>>
>>> If anyone wants more details, I'll be happy to share with you.
>>>
>>> -tom
>>>
>>>
>> I am considering using EC2 for a web application.
>> I am not sure how to calculate the payment per month.
>> Do I pay only for the time someone makes a request?
>> For example, I have a user who requests a certain report and it takes 1
>> second to load the report request form, then 20 seconds to produce the
>> report and print it.
>> I understand that I pay for 21 seconds?
>>
>
> In addition to mistakes already corrected, there is another mistake of how
> long something takes. Amazon aim to provide a certain computation power
> unit, but benchmarks show that what is actually provided has high
> variability. For example, ping times to EC2 machines started rising
> significantly since Amazon announces the spot instances. See also:
>
>

Some more input on EC2.

Not all instances born alike. We recently ran a huge computation based on
Hadoop and you can definitely see that some nodes perform faster (I/O was
the bottleneck) then others.

I too, when starting with EC2 made the mistake to of thinking that you only
pay for as much "CPU" as you use. Wrong!

OTOH, I was very happy to find out that with Google AppEngine this is
actually the case: You pay for as much resources as you consume. And they DO
count "CPU Time" vs. Amazon's "instance is running time".

Another note regarding EC2. Read bitbucket story about ec2 horrors
http://blog.bitbucket.org/.
Yet please don't get me wrong, generally EC2, S3, CloudFront, ELB and other
Amazon's services work great - Our production farm (~40 servers is hosted
there and we are relatively happy).

Amazon's main issues are:
I/O bandwidth is funny
Occasionally peaks in connectivity time that lead to timeouts (between zones
& from the outside world).
Not so fair hypervisor: We've seen occasions when an instance "slows down"
for a couple of minutes. We assume (without being able to tell for sure)
that some bigger instance type that happen to be hosted on the same physical
server as we are got resource hungry and practically ate all our CPU time...


Maxim.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.huji.ac.il/pipermail/linux-il/attachments/20101010/6c214dbe/attachment-0001.html>