Comments (3)
Thanks for the update.
We are currently testing a new detector that uses the safer pagemap interface instead of using /dev/mem. The new detector can be found in the repository: mc-mapping-pagemap.c.
The following is the result of the new detector on the Nehalem platform we used in the original PALLOC paper, which clearly shows bit 12,13,19,20 are used for the mapping. We plan to remove the old /dev/mem version in the future.
$ sudo chrt -f 1 ./mc-mapping-pagemap -p 0.7 -n 3
mem_size (MB): 2756
allocation complete.
worker thread begins
worker thread begins
worker thread begins
Bit6: 299.67 MB/s, 213.57 ns
Bit7: 307.61 MB/s, 208.05 ns
Bit8: 295.97 MB/s, 216.24 ns
Bit9: 297.84 MB/s, 214.88 ns
Bit10: 300.66 MB/s, 212.86 ns
Bit11: 245.89 MB/s, 260.28 ns
Bit12: 792.58 MB/s, 80.75 ns
Bit13: 789.23 MB/s, 81.09 ns
Bit14: 296.21 MB/s, 216.06 ns
Bit15: 294.19 MB/s, 217.55 ns
Bit16: 240.98 MB/s, 265.58 ns
Bit17: 294.20 MB/s, 217.54 ns
Bit18: 294.07 MB/s, 217.64 ns
Bit19: 789.05 MB/s, 81.11 ns
Bit20: 789.15 MB/s, 81.10 ns
Bit21: 294.17 MB/s, 217.56 ns
Bit22: 240.98 MB/s, 265.58 ns
Bit23: 294.10 MB/s, 217.62 ns
from palloc.
Hi, thanks for the information.
the new detector does not seem working well in my environment. I can provide further information if it would help improving the detector. The machine has only 1 DIMM and I set g_cache_num_ways
to match the actual associativity. Note that the machine is different from the one I used in #9.
I guess memory accesses generated by mc-mapping-pagemap
are served by the cache as their latency is reasonably shorter than the one measured by Intel Memory Latency Checker (65.1 ns).
$ cat /proc/cpuinfo | grep "model name" | uniq
model name : Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz
$ cat /sys/devices/system/cpu/cpu0/cache/index3/ways_of_associativity
20
$ sudo chrt -f 1 ./mc-mapping-pagemap -p 0.7 -n 3
mem_size (MB): 5518
allocation complete.
worker thread begins
worker thread begins
worker thread begins
Bit6: 1228.09 MB/s, 52.11 ns
Bit7: 1441.79 MB/s, 44.39 ns
Bit8: 1619.70 MB/s, 39.51 ns
Bit9: 1660.54 MB/s, 38.54 ns
Bit10: 1663.63 MB/s, 38.47 ns
Bit11: 1647.35 MB/s, 38.85 ns
Bit12: 1726.24 MB/s, 37.07 ns
Bit13: 1709.34 MB/s, 37.44 ns
Bit14: 1902.05 MB/s, 33.65 ns
Bit15: 1876.14 MB/s, 34.11 ns
Bit16: 1936.24 MB/s, 33.05 ns
Bit17: 1901.17 MB/s, 33.66 ns
Bit18: 1714.88 MB/s, 37.32 ns
Bit19: 1699.40 MB/s, 37.66 ns
Bit20: 1540.65 MB/s, 41.54 ns
Bit21: 1557.51 MB/s, 41.09 ns
Bit22: 1924.62 MB/s, 33.25 ns
Bit23: 1703.65 MB/s, 37.57 ns
$ sudo ./mlc
Intel(R) Memory Latency Checker - v3.1a
Measuring idle latencies (in ns)...
Memory node
Socket 0
0 65.1
from palloc.
For what it's worth, here are my results on an i7-6700HQ with two R2x8 DDR4 DIMMs (each with 16 banks and 4 bank groups):
$ sudo chrt -f 1 ./mc-mapping-pagemap -p 0.7 -n 3
mem_size (MB): 11136
allocation complete.
worker thread begins
worker thread begins
worker thread begins
Bit6: 4671.85 MB/s, 13.70 ns
Bit7: 4818.67 MB/s, 13.28 ns
Bit8: 4750.99 MB/s, 13.47 ns
Bit9: 4751.69 MB/s, 13.47 ns
Bit10: 4735.04 MB/s, 13.52 ns
Bit11: 4818.97 MB/s, 13.28 ns
Bit12: 4729.72 MB/s, 13.53 ns
Bit13: 4817.95 MB/s, 13.28 ns
Bit14: 4728.85 MB/s, 13.53 ns
Bit15: 4817.50 MB/s, 13.28 ns
Bit16: 1966.58 MB/s, 32.54 ns
Bit17: 4095.22 MB/s, 15.63 ns
Bit18: 3948.74 MB/s, 16.21 ns
Bit19: 3857.12 MB/s, 16.59 ns
Bit20: 3908.53 MB/s, 16.37 ns
Bit21: 1059.46 MB/s, 60.41 ns
Bit22: 918.14 MB/s, 69.71 ns
Bit23: 525.82 MB/s, 121.72 ns
I'm not sure how to interpret the results: the latencies for almost all bits seem way to low to reflect misses to main memory. Note that Skylake has a hashed L3, so you can't rely on simple ways/sets logic to force misses by overloading a way: you would need to know the hashing function (it has been reverse engineered elsewhere).
from palloc.
Related Issues (11)
- Issues with CGROUP when trying to apply cache partitioning HOT 1
- Some questions about mc-mapping.c
- detect-mc-mapping.sh fails on kernel 4.10 with segfaults
- ./detect-mc-mapping-pagemap.sh fails with invalid option -- 'b' HOT 2
- palloc on T1040D4RDB
- Kernel build fails when adding new parameters HOT 1
- Query: Any attempt to submit upstream? HOT 1
- Patch 3.13 HOT 1
- PALLOC usage for cache partitioning HOT 3
- Memory region used by mc-mapping 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 palloc.