Can You Capture Physical CPU-to-Memory Addresses in Real Time for Research?

  • Thread starter Thread starter Fubini
  • Start date Start date
  • Tags Tags
    Physical Research
Click For Summary

Discussion Overview

The discussion revolves around the feasibility of capturing physical CPU-to-memory addresses in real time for research purposes, particularly in the context of evaluating caching algorithms. Participants explore whether such data can be accessed publicly or generated through software on standard processors under real operating conditions.

Discussion Character

  • Exploratory, Technical explanation, Debate/contested

Main Points Raised

  • David expresses a need for real-time access to physical addresses for evaluating caching algorithms and inquires about public resources or software solutions.
  • One participant suggests that virtual addresses might be more relevant than physical addresses for the research.
  • Another participant notes that capturing internal cache hits and page hits is typically a hardware function, with the kernel only involved during specific memory access scenarios. They question whether virtual machines could provide this information, indicating it may not be in real time.
  • A different viewpoint mentions that using an ARM CPU with sufficient pinouts might allow tracing this information externally, suggesting a potential hardware solution.
  • One participant argues that the variability in cache sizes and performance across machines makes a study of that nature less useful, recommending simulation of hardware caches instead to evaluate performance on specific machines.

Areas of Agreement / Disagreement

Participants express differing views on the relevance of physical versus virtual addresses and the feasibility of capturing cache-related data, indicating that multiple competing perspectives exist without a clear consensus.

Contextual Notes

Limitations include the dependence on hardware capabilities for capturing cache data and the potential lack of real-time information from virtual machines. The discussion also highlights the variability in hardware configurations affecting the study's applicability.

Fubini
Messages
13
Reaction score
0
I'm doing some senior research into caching algorithms. In order to effectively evaluate some of our algorithms it would be amazing if we had access to physical addresses sent from the CPU to the memory captured in real time under real operating conditions. Does anyone know of a place that might have something like this in the public domain for research?

Alternately, would it hypothetically be possible to generate such a thing in software on a standard Intel processor running windows/linux whatever? Kernel hacking would not be out of the question.

Thanks,
David
 
Technology news on Phys.org
I think you want virtual addresses rather than physical ones
 
The issue with this is that it's done in hardware on most CPU's. The kernel only get's involved when there's an attempt to access a page that is not currently mapped in memory. I don't know if any CPU has a means to capture and/or report internal cache hits and/or page hits to some external device.

I don't know if running a virtual machine on an Intel or AMD cpu would provide this information. If so, that would be an alternative, although it would not be real time.

Perhaps something like a ARM cpu setup with sufficient pinouts to trace this information to an external device would be possible.
 
Last edited:
The size and performance of L1, L2 and L3 cache varies from machine to machine, so there is no use in doing a study of that. But if you want to evaluate the performance of a specific machine, the traditional way is to find out exactly how the hardware caches work, and then simulate them with software. That way, you can run all kinds of programs against the simulations and see what's happening.
 

Similar threads

Replies
29
Views
6K
  • · Replies 8 ·
Replies
8
Views
4K
  • · Replies 6 ·
Replies
6
Views
3K
  • · Replies 1 ·
Replies
1
Views
3K
Replies
8
Views
5K
  • · Replies 23 ·
Replies
23
Views
5K
Replies
6
Views
2K
  • · Replies 13 ·
Replies
13
Views
4K
  • · Replies 13 ·
Replies
13
Views
4K