First impressions of F1 Through 4

[Brandon Wirtz]

I wish I could get F .5 instances. And also F8’s J  

 

It doesn’t appear that memcache changes with this, so if you use Instance Memory in place of memcache we are seeing some HUGE performance increases on F4’s

At the same time, we have stuff optimized for memcache that we could  do a lot with Half sized instances a lot better.  Yeah concurrency, and all that.  We still see issues with the scheduler… I’d be quite content with 1/3 sized instances.  

It also appears that an F4 gets you a lot of bang for the buck because there is some per instance overhead otherwise.  We also see that spin up times are reduced on F4’s so you likely a lot of people who have 4 instances always could benefit from f4’s but the cost of those extra instances is high, especially since there isn’t a bulk purchase discount J

 

[Adam Sah]

We went with F2's for our frontends.  The cost was a bit high until I did a little tuning, which for our app ended up at 1-3 idle instances, pending latency 60ms – 14.9s

(we aggressively use cache-control: public on most things, so server latency isn't our problem)

[Adam Sah]

oops, forgot to mention.

Googlers-- if you're reading this, a supercool, googley feature: analyze the app's traffic and predict the $-impact of settings changes to idle instance and pending latency.

[Gregory D'alesandre]

We're listening :)  Definitely an interesting idea although pretty challenging to give an accurate estimate...

Greg

On Wed, Jan 4, 2012 at 3:36 PM, Adam Sah <as...@bbfdirect.com> wrote:

oops, forgot to mention.

Googlers-- if you're reading this, a supercool, googley feature: analyze the app's traffic and predict the $-impact of settings changes to idle instance and pending latency.

--
You received this message because you are subscribed to the Google Groups "Google App Engine" group.

To view this discussion on the web visit https://groups.google.com/d/msg/google-appengine/-/CT5YtjItjGgJ.

To post to this group, send email to google-a...@googlegroups.com.
To unsubscribe from this group, send email to google-appengi...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/google-appengine?hl=en.

[Brandon Wirtz]

The bigger your app the Better the savings… Some of the small apps that average 3 f1’s cost a LOT more on f2’s because the scheduler would spike to 4 instances when it only needed them briefly and over compensating suddenly cost twice as much.  (Lot is a % increase not a $ increase)

 

If you are running Python 2.7 and have 10 instances normally the advantage of F4’s is HUGE. We were blown away. The Larger Ram, and the faster handling resulted in SAVING 30% by running bigger instances.

 

I’m not sure that Google could build a tool that would deal with this at both ends of the spectrum.

--

[N. Rosencrantz]

I would like an investigation so I'm grateful for this thread.

F4 will bill me more and I don't know exactly how many instance hours I need in my case (100 daily at current levels?)

To make an estimate I think I want to know number of pageviews per instance hour and I can get that if I can complete a test with F4 for one complete day of billing and it's looking like I must pay > 4 dollars just to complete to test which is looking like I will go with F1 or F2. I tried changing from F1 to F4 and my billing skyrocketed from 0.1 to 4. I want to run my app on F4 but 4 dollars a day is expensive for a virtual server.

At current usage levels with F1 I'm billed abt 10 cents daily for my largest app while same time running it on F4 bills me > 4 dollars today so difference in billing is huge.

Maybe we could tell from the billing which billing was F1, F2 or F4 since I had to experimentally find that F2 should be my current setting.

I still couldn't try F4 for a complete day without going over quota.

In my latest test I set the daily budget to 4 USD and still couldn't run F4 without going over quota.

Increasing from F1 to F4 in my case more than quardruples the billing so without affecting income so for now I can't go with F4, it's too expensive and my app goes over quota. The difference in billing is not the difference in income. I can increase the budget but that won't increase income so I don't buy performance for its own sake.

Does the setting apply to static files and blobs as well or just the runtime? Now I want emil notification if my app goes over quota again, could that be enabled?

I use python 2.7, webapp2, jinja2, wtforms, babel.

Thanks,
Niklas

[Brandon Wirtz]

I doubt if you are in the free tier (or sub $1) That you need F4.

 

And $4 a day is CHEAP for a virtual server if it is done right.   We reduced several clients hosting for $3000-5000 a month to $120.  That’s a chunk of change for hosting.  And Sites like Jeff Probst’s went from needing $6k a month to handle bursts to needing $500 a year. That’s an insane reduction in cost.

--

You received this message because you are subscribed to the Google Groups "Google App Engine" group.

To view this discussion on the web visit https://groups.google.com/d/msg/google-appengine/-/xsQvFSo2h7wJ.