Roadmap of koordlet and slo-controller

[zwzhang0107]

Let's define the "priority one" topics for the next few milestones of koordlet and slo-controller. @stormgbs @jasonliu747 @saintube

[jasonliu747]

In my onw opinion, documentation is the number one priority of this project.

[zwzhang0107]

How about show a list of needed docs in your mind, and we can divide the work

[jasonliu747]

SGTM

[zwzhang0107]

Currently, several topics are important in my opinion.

1. Diversify the abilities of Koordlet, including a series of strategies about QoS Management, and collaboration with scheduler such as topology-aware scheduling
2. Well-designed modules with framework, which can be easily extended by other developers. Should start with QoS-Manager and RuntimeHooks.
3. Introduction docs, including designs and guides.

[zwzhang0107]

Now we have v0.3 already defined as following.

• cpu burst
• design doc of qos manager
• memory qos
• LLC and MBA isolation

[zwzhang0107]

last discussion, we have settle down the following issues to do

1. setting limits for extend-res - v0.4
2. improvement of memory evict - v0.4
3. QoS manager
   a. design - v0.4
   b. framework implement - v0.5
   c. fist plugins implement - v0.6
   c. more plugins implement - v0.7
4. runtime hook
   a. framework implement - v0.4
   b. first plugin implement - v0.5
   c. more plugins implement - v0.6
5. UT review
   a. cpu_suppress_test - v0.4
   b. memory_evict_test - v0.4
6. functional test scaffold - v0.5
7. docs (your list is expecting @jasonliu747), let's divide and conquer!

[zwzhang0107]

update:

1. setting limits for extend-res - v0.4
2. improvement of memory evict - v0.4
3. QoS manager
   a. design - v0.5
   b. framework implement - v0.5
   c. fist plugins implement - v0.6
   c. more plugins implement - v0.7
4. runtime hook
   a. framework implement - v0.4
   b. first plugin implement - v0.4
   c. more plugins implement - v0.6
5. UT review
   a. cpu_suppress_test - v0.4
   b. memory_evict_test - v0.4
6. functional test scaffold - v0.5
7. docs (your list is expecting @jasonliu747), let's divide and conquer!
8. other features
   a. cpu evict strategy - v0.4
   b. network qos - v0.4
   c. blkio qos - v0.5
   c. recommender - 0.6

[jasonliu747]

For a good project, there are several documents I think are essential:

1. Architecture
2. Quick Start Giude
3. Installing/Upgrading Koordinator
4. Best Practices
   • Memory QoS
   • Memory Evict
   • CPU Burst
   • CPU Supress
   • Deep Learning
   • Big Data
   • etc
5. FAQ
6. Troubleshooting

[zwzhang0107]

feature implement

• Memory QoS
• Memory Evict
• CPU Burst
• CPU Supress
• LLC/MBA
• cpu evict strategy
• network qos
• blkio qos

doc

• Memory QoS
• Memory Evict
• CPU Burst
• CPU Supress
• LLC/MBA
• cpu evict strategy
• network qos
• blkio qos

[zwzhang0107]

milestone v0.5:
○ support runtime-plugin bypass running mode (#120 )
○ qos-manager design (#58)
○ support batch-res limit != request (#83)
○ pqos，support llc/mba for amd (maybe doing some research first) (#216)
○ compatible internal design