evaluate NUMA memory binding

[psychocrypt]

Based on the discussion in #108 it could be useful to evaluate if there is any improvemnt if we can bind memory to the correct NUMA node.

@fireice-uk's observation

One of the points I noted when testing on multi-cpu, multi-ram systems (on Linux) is that the results varied by around 5% but were constant during the run, and usually during a run shortly after, whereas they were most likely to change after reboot.

[psychocrypt]

@fireice-uk Could you please post the hwloc output hwloc-ls --output-format ascii from your laptop.

[fireice-uk]

┌─────────────────────────────────────────────────────────────────────────────┐
│ Machine (7690MB)                                                            │
│                                                                             │
│ ┌────────────────────────────────┐            ┌───────────────────────────┐ │
│ │ Package P#0                    │  ├┤╶─┬─────┤ PCI 8086:0a16             │ │
│ │                                │      │     │                           │ │
│ │ ┌────────────────────────────┐ │      │     │ ┌───────┐  ┌────────────┐ │ │
│ │ │ L3 (3072KB)                │ │      │     │ │ card0 │  │ renderD128 │ │ │
│ │ └────────────────────────────┘ │      │     │ └───────┘  └────────────┘ │ │
│ │                                │      │     │                           │ │
│ │ ┌────────────┐  ┌────────────┐ │      │     │ ┌────────────┐            │ │
│ │ │ L2 (256KB) │  │ L2 (256KB) │ │      │     │ │ controlD64 │            │ │
│ │ └────────────┘  └────────────┘ │      │     │ └────────────┘            │ │
│ │                                │      │     └───────────────────────────┘ │
│ │ ┌────────────┐  ┌────────────┐ │      │                                   │
│ │ │ L1d (32KB) │  │ L1d (32KB) │ │      │     ┌───────────────┐             │
│ │ └────────────┘  └────────────┘ │      ├─────┤ PCI 8086:1559 │             │
│ │                                │      │     │               │             │
│ │ ┌────────────┐  ┌────────────┐ │      │     │ ┌─────────┐   │             │
│ │ │ L1i (32KB) │  │ L1i (32KB) │ │      │     │ │ enp0s25 │   │             │
│ │ └────────────┘  └────────────┘ │      │     │ └─────────┘   │             │
│ │                                │      │     └───────────────┘             │
│ │ ┌────────────┐  ┌────────────┐ │      │                                   │
│ │ │ Core P#0   │  │ Core P#1   │ │      │               ┌───────────────┐   │
│ │ │            │  │            │ │      ├─────┼┤╶───────┤ PCI 8086:08b2 │   │
│ │ │ ┌────────┐ │  │ ┌────────┐ │ │      │               │               │   │
│ │ │ │ PU P#0 │ │  │ │ PU P#2 │ │ │      │               │ ┌────────┐    │   │
│ │ │ └────────┘ │  │ └────────┘ │ │      │               │ │ wlp3s0 │    │   │
│ │ │ ┌────────┐ │  │ ┌────────┐ │ │      │               │ └────────┘    │   │
│ │ │ │ PU P#1 │ │  │ │ PU P#3 │ │ │      │               └───────────────┘   │
│ │ │ └────────┘ │  │ └────────┘ │ │      │                                   │
│ │ └────────────┘  └────────────┘ │      │     ┌───────────────┐             │
│ └────────────────────────────────┘      └─────┤ PCI 8086:9c03 │             │
│                                               │               │             │
│                                               │ ┌─────┐       │             │
│                                               │ │ sda │       │             │
│                                               │ └─────┘       │             │
│                                               └───────────────┘             │
└─────────────────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────────────────┐
│ Host: LAP440                                                                │
│                                                                             │
│ Indexes: physical                                                           │
│                                                                             │
│ Date: Sun 14 May 2017 21:35:42 BST                                          │
└─────────────────────────────────────────────────────────────────────────────┘

Not sure if the ascii art will get garbled.

[psychocrypt]

Could you please try to use the newest version of xmr-stak with hwloc? It looks like core 0 and 2 are the physicle cores in your system( not 0,1). The new code with the function to overload a system if 1MiB for a hash is avalable should also suggest to use core 0 and 2. This should remove the fluctuations in the hashrate.

I also used always core 0,1 until hwloc shows that 0,2 are the physical cores.

[fireice-uk]

I will give it a spin, and yes, seems you are correct on that point, we will need to properly allocate core affinity it seems:

cat /proc/cpuinfo | grep 'processor\|core id\|physical id'
processor	: 0
physical id	: 0
core id		: 0
processor	: 1
physical id	: 0
core id		: 0
processor	: 2
physical id	: 0
core id		: 1
processor	: 3
physical id	: 0
core id		: 1

[psychocrypt]

@fireice-uk information: I am currently working on this issue. I have finished a prototype where the memory for a thread should be pinned to the numa node.
At the moment I can't see any benefit under linux. I will clean my code and create a pull request soon.

[fireice-uk]

Ok. Great, what's the setup you are testing on?