Is anyone else surprised the EC2 Micro did so well?
If you read the EC2 forums[1] for any amount of time you get used to seeing post after post of "hung sites" running on the Micros when a constant level of demand is made on them (not the intended usage model[2])
Under load the Micros frequently go into a catatonic state with %st ('top') climbing to the moon as the VM environment provisions the brunt of the server's resources to the other VMs on the machine and starves out any hungry Micro instances (as designed).
I didn't think you could host anything on them with regularity. Maybe a swarm of them behind an ELB, not not a single one... anyone else had the same experience that Malda had?
Last time I tried the micro's they were mostly starving on I/O.
Quite possibly his webserver wasn't doing much of that?
Also "tens of thousands" of requests is not really a big number anymore. Nginx will happily serve a couple thousand per second even from a micro (until it gets choked by the EC2 resource beancounting).
moe that's a good point, if he's got all the pages statically cached and all the graphical assets on S3, all nginx would need to do is serve back relatively small TXT files which even on the micro would be trivial.
On micros, if the load is CPU intensive, the vCPU gets throttled pretty quickly for sustained loads (say over 10 - 15 seconds) and your Steal Time CPU percentage goes up -the hypervisor will give someone else real CPU cycles before coming back to serve you. So, they are good for low traffic sites with bursty loads but not good for sustained traffic/loads.
That is a higher traffic site than I expected a Micro to be able to handle.
Are you hosting dynamic content (PHP, Java, etc.) or is everything really heavily cached in static files/varnish and hosted out by nginx quickly so as to avoid saturating CPU or I/O?
Static. On this site I host mostly huge-ass html files (about 1 MB per request) with apache 2 and mod_deflate :-)
There is also a wordpress site and another database-request-heavy site with 500 requests a day.
The problems I had was that the access.log and error.log quickly ate up the 5GB space of the micro instance :-) Causing mysql to stop operating and me to scratch my head for a few minutes.
5GB is not a lot .. you want ruby1.9 with the default setup? That will be 180MB, and so on :-)
If you read the EC2 forums[1] for any amount of time you get used to seeing post after post of "hung sites" running on the Micros when a constant level of demand is made on them (not the intended usage model[2])
Under load the Micros frequently go into a catatonic state with %st ('top') climbing to the moon as the VM environment provisions the brunt of the server's resources to the other VMs on the machine and starves out any hungry Micro instances (as designed).
I didn't think you could host anything on them with regularity. Maybe a swarm of them behind an ELB, not not a single one... anyone else had the same experience that Malda had?
[1] https://forums.aws.amazon.com/thread.jspa?threadID=58323
[2] http://docs.amazonwebservices.com/AWSEC2/latest/UserGuide/in...