For hbase, we need to reduce the number of minihbasecluster start/stop.

Each James independent tests suppose a clean persistent store, this is why we need to start from scratch (easy for memory and file, takes more time for other persistence technologies).

btw, this is also a topic on which the hbase project works for its own tests (running a full test suite on hbase src takes hours, mainly due to those numerous minihbasecluster start/stop).


Thx,
Eric


On 16/10/11 17:14, Manuel Carrasco Moñino wrote:
Thanks Stefano

I just run once each test.

I know there is a overhead when running setup and teardown, what I do
not know is if the test could be useful for performance, subtracting
the setup/teardown times and looping it.

- Manolo

On Sun, Oct 16, 2011 at 5:05 PM, Stefano Bagnara<[email protected]>  wrote:
2011/10/16 Manuel Carrasco Moñino<[email protected]>:
Hi all,

Is the Manager Test useful to measure performance?

Testing the the couchdb implementation Pepijn is working on, I've
released that some implementations takes a long to run the tests (see
attached screenshot):
InMemory 0.7s
MailDir 2.4s
JPA 6.0s
JCR 21.1s
CouchDB 40.4s
HBase 81.8s

JPA uses a ddbb in memory, CouchDB is not optimized at all, but Is
there any reason why jcr and hbase takes so long?

Don't know how you test this, but if you include startup and warmup
times then it is expected that more complex layers will take longer to
startup.
Maybe you should run the tests 1000 times or make sure you at least
run them twice and only measure timings for the second time (and
exclude startup times).

Does this make sense?

Stefano

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]



---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]


--
Eric
http://about.echarles.net

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to