Mixed workload paramters clarification
Badb0yBadb0y opened this issue · comments
Hello,
It's not really detailed what the following parameters means in the mixed workload so it's difficult to generate my workload:
The distribution of operations can be adjusted with the --get-distrib, --stat-distrib, --put-distrib and --delete-distrib parameters.
I'd like to mimic this with the tests:
PUT: 9379 47% PUT
GET: 7111 36% get
DELETE: 3410 17% delete
19900 altogether
Tried with 2 different way
- Defined as a percentage that I want to achieve:
warp mixed --noclear --put-distrib=47 --get-distrib=36 --delete-distrib=17 --duration=5h --objects=1000000 --obj.size=40k --bucket=2andhalfbillion --host=... - Defined as a calculation based on the default 20 concurrency, so if the below values multiplied with 20, I should get that number that I want
warp mixed --noclear --put-distrib=469 --get-distrib=356 --delete-distrib=171 --duration=5h --objects=1000000 --obj.size=40k --bucket=2andhalfbillion --host=...
but still not proper numbers.
If you can see my graph, it's a way more higher (take into consideration only the delete and the get because there is a ut operation running next to those)
Graph
How is the proper way to mimic my workload?
You must set the --stat-distrib=0
, otherwise it will use the default value on 30.
warp mixed --put-distrib=47 --get-distrib=36 --delete-distrib=17 --stat-distrib=0 ...
Without setting, the distribution will be ~
op: GET dist: 27.7%
op: STAT dist: 23.1%
op: PUT dist: 36.2%
op: DELETE dist: 13.1%
This does not include the initial 1M PUT requests.
Your graph doesn't really make sense to me. If I read it correctly you have something else running while the benchmark is running. Hard to tell.
Maybe try shorter runs until you get the settings as you want, then do a longer with more initial objects.