Comments (4)
@ivancich could you help to take a look ?
from dmclock.
You're correct that the issue likely involves the calculation of rho and delta. The nature of the algorithm means the limit can never be perfectly enforced. Rho and delta are calculated by the client to give a snapshot summary of the recent service it has received from other servers. So when there's no recent history a client can get lots of service distributed over the servers, and then there will be a compensation afterward. That can lead to cycles of over- and under-service. Perhaps the algorithm can be modified to dampen such cycles over time.
You seem to be getting more dramatic cycling than I get. Here's a run with your same configuration file:
==== Client Data ====
client: 0 1 2 97 98 99 total
t_0: 54.50 53.00 53.50 53.50 51.00 53.50 5333.50
t_1: 34.50 34.50 36.50 33.50 35.50 36.50 3658.50
t_2: 30.00 32.00 30.00 31.50 31.50 28.00 3016.50
t_3: 30.00 30.00 29.50 30.00 29.00 30.50 2990.50
t_4: 31.00 30.00 30.00 30.50 30.50 29.50 2993.00
t_5: 31.50 30.00 30.00 32.50 29.50 30.50 3005.50
t_6: 29.00 30.00 30.00 28.00 30.00 30.00 3013.50
t_7: 30.00 30.00 30.00 29.00 30.00 29.00 2987.00
t_8: 30.00 30.00 30.00 30.00 30.00 31.00 3013.00
t_9: 30.00 30.00 30.00 30.00 30.00 30.00 2998.50
t_10: 30.00 30.00 30.00 30.50 30.00 29.50 3000.00
t_11: 30.00 30.00 30.00 31.00 30.00 30.50 2999.00
t_12: 30.00 30.00 30.00 29.50 30.00 29.50 2992.00
t_13: 30.00 30.00 30.00 29.00 30.00 30.00 2999.00
t_14: 29.50 29.50 32.00 29.50 30.00 37.00 3074.50
t_15: 20.00 21.00 18.50 22.00 23.00 15.00 1919.00
t_16: 0.00 0.00 0.00 0.00 0.00 0.00 7.00
t_17: 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Since you have server_random_selection set to false, each client sends its requests to its 10 servers in a very orderly manner -- a, b, c, d, e, f, g, h, i, j, a, b, c, d, e, f, g, h, i, j, .... When I take your configuration and set server_random_selection to true, the service becomes much less consistent, but every client seems to be over-served.
==== Client Data ====
client: 0 1 2 97 98 99 total
t_0: 57.50 50.50 56.50 57.50 55.50 53.50 5668.00
t_1: 41.00 52.00 44.00 46.50 45.00 45.50 4653.00
t_2: 36.50 35.50 43.00 41.00 41.00 44.50 3944.00
t_3: 42.50 27.50 40.50 38.50 39.50 39.00 3873.00
t_4: 37.50 47.00 37.50 38.50 33.50 42.50 3808.50
t_5: 40.50 39.00 44.50 37.00 40.50 34.00 3851.50
t_6: 32.00 39.50 28.50 40.50 38.50 43.00 3786.50
t_7: 33.00 38.50 34.50 36.00 47.50 44.50 3791.00
t_8: 39.50 36.50 40.50 35.50 34.50 38.50 3828.00
t_9: 40.50 37.00 41.00 34.50 41.50 35.00 3816.00
t_10: 47.00 40.50 37.00 36.00 38.00 34.00 3804.50
t_11: 32.00 39.00 20.00 36.00 25.50 29.50 3190.00
t_12: 14.50 9.00 13.50 11.50 13.00 10.50 1144.00
t_13: 2.50 4.00 11.50 4.50 6.00 3.00 441.50
t_14: 1.00 2.50 4.00 3.00 0.50 1.50 197.00
t_15: 1.50 2.00 2.00 2.00 0.00 1.50 93.50
t_16: 1.00 0.00 0.50 1.50 0.00 0.00 42.00
t_17: 0.00 0.00 1.00 0.00 0.00 0.00 25.00
t_18: 0.00 0.00 0.00 0.00 0.00 0.00 14.50
t_19: 0.00 0.00 0.00 0.00 0.00 0.00 9.00
t_20: 0.00 0.00 0.00 0.00 0.00 0.00 8.00
t_21: 0.00 0.00 0.00 0.00 0.00 0.00 5.00
t_22: 0.00 0.00 0.00 0.00 0.00 0.00 2.50
t_23: 0.00 0.00 0.00 0.00 0.00 0.00 2.00
t_24: 0.00 0.00 0.00 0.00 0.00 0.00 2.00
t_25: 0.00 0.00 0.00 0.00 0.00 0.00 0.00
So given the total column, it seems as though on average each client is receiving around 38 reqs/sec, which is about 27% more than we'd ideally expect. I will review the calculation of rho and delta and see if I can detect an issue.
Thank you.
from dmclock.
Hi Eric, how's it going.
Would you mind telling me the plan of dmclock in ceph, and when does the Qos feature can be used?
from dmclock.
The first version that will be integrated will be focused on modifying the balance of operations between client ops, osd subops, recover, snaptrim, and scrub. From there we'd like to provide even more flexibility.
from dmclock.
Related Issues (13)
- When can we use dmclock in Ceph? HOT 6
- Why "dmclock" is wrappered under the namespace of "crimson"? HOT 2
- confused error while compiling "dmclock-tests" HOT 3
- where is gtest/gtest_prod.h HOT 6
- Segmentfault encountered during rados/basic testing HOT 5
- There are "out of order op" errors while testing qa suites HOT 20
- What's the license? HOT 3
- how to configure the io's r, w, l parameters? HOT 5
- how to use dmclock in multi-thread io model? HOT 6
- How to config mclock_client queue in ceph mimic? How to distinguish different clients? HOT 5
- will dmclock queue get stuck?
- how to configure reject_threshold HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from dmclock.