memory locking on mac!

[Cathrach]

Ticket(s) Closed

• Closes Implement memory locking (to avoid memory swap to disk) #6923

Description
This PR adds code to use patched mimalloc as a substitute for system malloc, so we can mlock our allocations.
It also includes a function to mlock statics, which is ad-hoc and requires more testing/possibly a better method. I am extremely open to suggestions.

See https://www.notion.so/whisthq/Mimalloc-and-mlock-Implementation-c2e7d80ce8dc406981928bd0e4560b89.

Implementation

Documentation & Tests Added

Testing Instructions

Pull whisthq/mimalloc, check out the serina/mlock_error branch, and build that using the building instructions. Copy the mimalloc.o file to the build/lib/etc/mimalloc/etc folder on the protocol side

On the protocol side:

• make sure you compile with -D USE_MIMALLOC=ON
• To verify that mlocking is working as intended, you can turn on MLOCK_LOG in the mimalloc side; you can also check that Whist's memory usage is higher than dev Whist
• To check that this is actually improving performance:
  • turn on PLOT_UDP_RECV_GAP in debug_flags.h
  • turn on LOG_DATA_FOR_PLOTTER in log_statistics.h and change other settings as you wish
  • stress your memory
  • use plotter.py

Serina's ABABABAB test (suggested by Yancey)
with mlock:

dev:

with mlock:

dev:

with mlock:

dev:

with mlock:

dev:

PR Checklist

• Did the PR author fully test this PR end-to-end?
• Did one PR reviewer fully test this PR end-to-end?
• Did one PR reviewer conduct a thorough code design review?

[wangyu-]

Is there a link to some data of this PR vs dev?

Seems like the only data so far is a screenshot in the protocol channel.

I suggest do rigorous test/experiment, and post more data in the PR description. (As I mentioned in #7011)

[philippemnoel]

is the patch file still needed?

[Cathrach]

Is there a link to some data of this PR vs dev?

Seems like the only data so far is a screenshot in the protocol channel.

I suggest do rigorous test/experiment, and post more data in the PR description. (As I mentioned in #7011)

I will do this over the weekend and post images! yeet

[philippemnoel]

Could we solve the TODOs?

[codecov]

Codecov Report

Merging #7373 (b159538) into dev (c104c46) will increase coverage by 1.25%.
The diff coverage is n/a.

❗ Current head b159538 differs from pull request most recent head 3a9e3c9. Consider uploading reports for the commit 3a9e3c9 to get more accurate results

@@            Coverage Diff             @@
##              dev    #7373      +/-   ##
==========================================
+ Coverage   56.13%   57.38%   +1.25%     
==========================================
  Files         158      106      -52     
  Lines       32519    26947    -5572     
==========================================
- Hits        18255    15464    -2791     
+ Misses      14010    11483    -2527     
+ Partials      254        0     -254     

Flag	Coverage Δ
backend-services	?
protocol	57.38% <ø> (-0.16%)	⬇️

Impacted Files	Coverage Δ
protocol/client/whist_client.cpp	0.94% <ø> (ø)
protocol/whist/core/whist_memory.c	68.30% <ø> (ø)
protocol/whist/fec/wirehair/WirehairCodec.cpp	88.21% <0.00%> (-2.55%)	⬇️
backend/services/host-service/auth/auth.go
backend/services/metadata/providers.go
backend/services/host-service/host-service.go
...ling-service/scaling_algorithms/default/default.go
...d/services/scaling-service/dbclient/mandelboxes.go
...aling-service/scaling_algorithms/default/assign.go
...aling-service/scaling_algorithms/default/deploy.go
... and 47 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7175fb1...3a9e3c9. Read the comment docs.

handled

[github-actions]

Protocol End-to-End Streaming Test Results

Experiments Summary

Expand Summary

✅ Experiment 1 - Bandwidth: Unbounded, Delay: None, Packet Drops: None, Queue limit: default. Download logs:

aws s3 cp s3://whist-e2e-protocol-test-logs/serina/mimalloc/2022_11_21@19-43-12/ 2022_11_21@19-43-12/ --recursive

✅ Experiment 2 - Bandwidth: variable between 15Mbit and 30Mbit, Delay: 10 ms, Packet Drops: None, Queue Limit: None, Conditions change over time? Yes, frequency is variable between 1000 ms and 2000 ms. Download logs:

aws s3 cp s3://whist-e2e-protocol-test-logs/serina/mimalloc/2022_11_21@19-49-30/ 2022_11_21@19-49-30/ --recursive

✅ Experiment 3 - Bandwidth: variable between 10Mbit and 20Mbit, Delay: 10 ms, Packet Drops: None, Queue Limit: None, Conditions change over time? Yes, frequency is variable between 500 ms and 2000 ms. Download logs:

aws s3 cp s3://whist-e2e-protocol-test-logs/serina/mimalloc/2022_11_21@20-07-55/ 2022_11_21@20-07-55/ --recursive

✅ Experiment 4 - Bandwidth: 10Mbit, Delay: 10 ms, Packet Drops: None, Queue Limit: 100 packets, Conditions change over time? No.. Download logs:

aws s3 cp s3://whist-e2e-protocol-test-logs/serina/mimalloc/2022_11_21@20-12-11/ 2022_11_21@20-12-11/ --recursive

Full Results: link here