How Amazon Followed Google Into the World of Secret Servers (2012)(wired.com)
wired.com
How Amazon Followed Google Into the World of Secret Servers (2012)
http://www.wired.com/2012/11/amazon-google-secret-servers/
2 comments
I read this article and see a future in which Google, Amazon and Goldman are at full stop because a bulk of these machines shipped with an embedded killswitch in the NIC or motherboard. I hope a random sample is being tested with regularity.
"As University of California, Berkeley, computer science professor David Patterson points out, Amazon Web Services were remarkably cheap from the very beginning, and the prices continue to drop."
Errmm.. no?
Errmm.. no?
I was thinking that myself. I have been using it since 2010, and prices since then are down by seemingly 50% and still expensive - so in the beginning, no it was not cheap. The IO is both bad and unpredictable, and even high costs instances are highly variable in performance on the same instance cost. meanwhile, their numbers are far bellow that of a hard machine you own, or get from a rackspace company.
they do save a lot of money if you do not need to hire certain types of IT fokes, and aws is extremely cost effective for satisfying the need when you have low traffic and then sudden large bursts of it that must be handled.
they do save a lot of money if you do not need to hire certain types of IT fokes, and aws is extremely cost effective for satisfying the need when you have low traffic and then sudden large bursts of it that must be handled.
Are you talking about Elastic Block Store IO? Move away from original EBS volumes, use General Purpose SSSD (now default) or provisioned IOPS volumes.
What do you recommend for hosting?
I've been looking at vultr alongside a Linode account for development and small website hosting purposes that need a little more customization and server side control than something like a Jekyll blog. AWS S3 is fairly reasonable for hosting small, static sites that get almost no hits besides you maybe ahead of almost everyone because most will charge you for the compute and memory of the web server of the site hosting the content while with it's stripped to transfer and static disk storage, which would be pennies per year for sharing a small collection of pictures with a few friends. I have gigabytes of stuff and it costs me less than a dollar per month on S3. Expensive if you look at Google Drive or OneDrive costs sure, but the utility I get out of it is worth a few dollars per year for me. Also, I like the security I can tweak on my bucketa.
With the HA design you tend to need running on AWS for serious operations (AZs out the wazoo with every other feature tacking on a few dollars at a time), you might as well use multiple cloud hosting providers and load balance across them. Obviously this isn't typical practice, but Amazon's lock-in is pretty glaring when looking at one of the best potential aspects of cloud being that of PORTABLE workloads.
With the HA design you tend to need running on AWS for serious operations (AZs out the wazoo with every other feature tacking on a few dollars at a time), you might as well use multiple cloud hosting providers and load balance across them. Obviously this isn't typical practice, but Amazon's lock-in is pretty glaring when looking at one of the best potential aspects of cloud being that of PORTABLE workloads.
i cant answer that, its highly dependent on your own needs.
I would love others to weigh in, correct me, add but I'm assuming not your own infrastructure, which is a valid option :
for a small webpage/server, perhaps linode or digital ocean. I find heroku nice for free hosting, but they are a bit pricy for paid services. I do love how easy it is to push code to heroku (especially with ruby on rails). if you never used amazon and only want to use it for a year, their micro instance is free for one year.
for large scaling systems and a suite of services, I would suggest AWS. they have so many services it confuses the heck out of me sometimes. if you want low cost high performance, especially for development digital ocean seems better for this.
i never used Azure, cant comment on it. Also there was a company that basically rented physical servers at a cheap price, from what I remember they had the best performance per cost that I have seen. Also another obvious consideration is software/applications - some tie in better than others. for instance, I use ArcGIS Server which is VERY easy to deploy on AWS
for a small webpage/server, perhaps linode or digital ocean. I find heroku nice for free hosting, but they are a bit pricy for paid services. I do love how easy it is to push code to heroku (especially with ruby on rails). if you never used amazon and only want to use it for a year, their micro instance is free for one year.
for large scaling systems and a suite of services, I would suggest AWS. they have so many services it confuses the heck out of me sometimes. if you want low cost high performance, especially for development digital ocean seems better for this.
i never used Azure, cant comment on it. Also there was a company that basically rented physical servers at a cheap price, from what I remember they had the best performance per cost that I have seen. Also another obvious consideration is software/applications - some tie in better than others. for instance, I use ArcGIS Server which is VERY easy to deploy on AWS
Maybe the true genius of AWS is the pricing model that makes expensive look cheap.
The beauty of AWS is that the marginal cost[1] is low.
It might cost slighly more to use an AWS instance than servers(s) than to run server(s) of similar computing power. But once you build those servers, you're stuck with them. You also need a swathe of IT staff to care for them. Thus, the marginal cost of adding computing power is quite high, since it's hard to scale back, short of selling the server or leasing out it's capacity to others.
AWS was one of the first _usable_ services that allowed people to purchase computing power on demand easily.
[1] https://en.wikipedia.org/wiki/Marginal_cost
It might cost slighly more to use an AWS instance than servers(s) than to run server(s) of similar computing power. But once you build those servers, you're stuck with them. You also need a swathe of IT staff to care for them. Thus, the marginal cost of adding computing power is quite high, since it's hard to scale back, short of selling the server or leasing out it's capacity to others.
AWS was one of the first _usable_ services that allowed people to purchase computing power on demand easily.
[1] https://en.wikipedia.org/wiki/Marginal_cost