Followup:

I contacted AppFog support via their live chat to get a few questions 
answered:

   - Confirmed: You can't buy more storage (yet). It's a priority for them, 
   though.
   - An app's instances don't all see the same copy of the filesystem. 
   That's less of a problem than the fact that:
   - Your writable filesystem isn't persistent. AppFog recommend it be 
   devoted strictly to your codebase, with any static assets being placed on 
   something like S3.
   - Confirmed: They don't provide cron. I attempted to use web2py cron in 
   my AppFog app, but I can't make it work. I haven't tried very hard to 
   figure out why.

I was also able to get my app running on their Rackspace instances- it 
seems their `af` command doesn't work right with the --debug flag, which I 
was using to see why things weren't working :P `af start <app>` worked fine.

Lastly, the Apache Benchmark numbers I presented last time lasted through 
at least Monday night. When I tested tonight, they had fallen to around 1 
second average load time- far more reasonable. I asked AppFog support about 
the high load times last weekend, and they couldn't pinpoint a specific 
issue, but said they had been fiddling with things in response to opening 
to the public and I might have just caught them at a bad time. I'll give 
them the benefit of the doubt on that, but I'd still keep an eye on those 
load times for a while before assuming an app will perform reasonably on 
their service. 

So, now that I got a Rackspace deployment working, and their hosting isn't 
slower than molasses, I'm a little more optimistic about AppFog. 


On Sunday, July 29, 2012 12:04:06 AM UTC-4, spiffytech wrote:
>
> Tonight I threw up a copy of my personal site, just to see if AppFog is 
> worth looking into. Perhaps my experience would have been better if I tried 
> during business hours while their live chat support was open, but I don't 
> plan to use AppFog after this experience.
>
> The highlights:
>
>    - They advertise unlimited apps, but you're limited by how many apps 
>    you can squeeze into the RAM your account gets. Sure, you don't have the 
>    hard cap of 10 apps like App Engine (still?) gives you, but "unlimited" 
>    isn't really true
>    - AppFog's founder wants you to 
> believe<http://blog.appfog.com/if-paas-is-expensive-and-slow-why-not-use-a-vps/>they're
>  proving PaaS doesn't have to be slow and expensive, but I found 
>    AppFog to be slow, and their pricing<https://console.appfog.com/pricing> 
> gets 
>    nutty-expensive very fast
>    - Getting my app working at all was a trying experience, and AppFog 
>    doesn't offer much in the way of documentation or debug output to help you
>    - No sign of a cron system, so you'll be relying on web2py's built-in 
>    cron (didn't check if it works there, but I assume so)
>    - All apps have a 100MB disk limit, and I don't see a way to buy more. 
>    Better hope you're apps don't get very big!
>    - You do have a writable filesystem, for what that's worth with the 
>    100MB disk use limit. I didn't check whether all instances access the same 
>    FS (that's kind of an important way so design the service)
>
> My conclusion: If you want a free place to host something that (really, 
> really) doesn't need to be performant, AppFog is a decent choice because:
>
>    - It has a writable filesystem, which sets it apart from App Engine
>    - It has free MySQL (App Engine started charging for MySQL in June), 
>    which sets it apart from App Engine
>    - It looks more likely to stick around than some of the other free web 
>    host services that I've seen mentioned here
>
> However, to get that free hosting you'll have to put up with terrible 
> performance (or highly variable performance if it magically speeds up by 
> tomorrow morning), poor documentation, and a tricky and opaque setup 
> procedure.
>
> On to doing stuff, and statistics!
>
> First off, I could not find any links on their site instructing me on how 
> to configure a Python app to work on AppFog. I eventually gave up and 
> resorted to Googling for a tutorial, which led me to this 
> section<http://docs.appfog.com/frameworks/python>in AppFog's docs. Not sure 
> how you are supposed to find that.
>
> That link isn't too helpful, though- it shows how to make a Flask site 
> that works on AppFog, and links to working Bottle and Django sites, but 
> doesn't spell out how to make a generic WSGI site work. To make my simple 
> web2py site work, I had to do the following:
>
> mv wsgihandler.py wsgi.py  # AppFog needs wsgi.py. I tried a symlink 
> instead of a move, but couldn't make AppFog work in that arrangement
> ln -s wsgi.py wsgihandler.py  # This ensures updates to web2py affect 
> your wsgi.py
> gem install af
> af login
> af push <appname>  # This gives you an Amazon East app. I can't figure 
> out how to use `af` to deploy to a different infrastructure
>
> I tried creating an app on the Rackspace infrastructure through AppFog's 
> web admin `af update <appname>`, but couldn't get my app to start. It 
> didn't start automatically, and `af start <appname> --debug` tells me I 
> don't have the "run" mode available. So no Rackspace for me.
>
> Once I got my app running on AppFog's Amazon EC2 infrastructure (after a 
> number of false starts related to not having wsgi.py) I noticed my app ran 
> very slowly. The front page of my app doesn't really do anything; it could 
> almost be a static HTML file, yet it was unbearably slow on AppFog.
>
> I fired up Apache Benchmark and got some very disappointing results. With 
> a concurrency of 50, run for 30 seconds: 
>
>    - 1 instance, 128MB RAM: 139 requests completed, mean average of 10.5 
>    seconds to fulfill a request
>    - 1 instance, 2GB RAM: 140 requests, 10.8 seconds mean
>    - 15 instances, 128MB RAM each: 163 requests, 9.2 seconds mean
>
> For comparison, my site hosted on its usual low-end rackmount server (RHEL 
> 6.2, 3.1GHz quad-core Xeon, 8GB RAM, Apache with mod_wsgi) completed 3061 
> requests with a mean 0.490 seconds per request. Beefier than what AppFog 
> gave me? Sure, but not enough to explain handling 22x the requests, with 
> 1/22nd the response time.
>
> *  The AppFog mean request fulfillment numbers varied by as much as two 
> seconds over the several times I ran the tests. That sort of 
> unpredictability worries me.
>
>
>
> On Thursday, July 26, 2012 9:50:01 PM UTC-4, Joel Carrier wrote:
>>
>> Has anyone tried running web2py on appfog ( www.appfog.com ) and cares 
>> to comment on their experience?
>
>

-- 



Reply via email to