Hi,
I changed my test so it sends a batch of requests every 2 seconds.
The batch size starts out as 1 and increases by 1 every x seconds.
Last run I did I had it increase by 1 every 100 seconds. This ramped
up things slowly.
I still started getting errors around 16-20 requests per second. My
additionally, under this load, the time it takes to service a request
grows from 2 seconds per request (no load) to 15 seconds per request
(high load).
On Apr 14, 11:07 pm, Anonymous Coderrr greedw...@gmail.com wrote:
I have a fairly simple app - it looks up a couple of objects from the
Hi,
Since this is just a test app, can you post the code (perhaps to Pastie
[1], for syntax coloring and the like) and your mechanism for storing
the data? That might help people help you figure out why it isn't
performing as you would hope.
[1] http://pastie.org
FWIW,
--
T.J. Crowder
tj /
One thing that has become apprent is appengine, is designed to scale
under real world usage.
So if your App went from 0/1 users to 500 in a matter of seconds, then
the system wont work well. You need to ramp up the usage slowly.
Even a slashdotting would result in a 'ramp' usage.
Also 500
Good points.
I rewrote the test so it fires off 20 requests from 20 distinct ip
addresses simultaneously, once a second 10 times.
In effect 20 concurrent requests, once a second.
I had about 15% loss and the request time degradation was there. (2
seconds to fulfill a request on an idle
You need to ramp up your tests over many minutes to allow GAE to spawn
enough instances to handle the traffic. I don't know their exact
algorithm, but I know it takes time to scale up.
On Apr 15, 1:33 pm, Anonymous Coderrr greedw...@gmail.com wrote:
Good points.
I rewrote the test so it