1 |
J. Roeleveld wrote: |
2 |
> On 4 November 2015 13:14:18 CET, hw <hw@×××××.de> wrote: |
3 |
>> |
4 |
>> Hi, |
5 |
>> |
6 |
>> how do I know whether xen uses NUMA or not? It says in dmesg: |
7 |
>> |
8 |
>> |
9 |
>> [ 0.000000] NUMA turned off |
10 |
>> [ 0.000000] Faking a node at [mem |
11 |
>> 0x0000000000000000-0x0000000040068fff] |
12 |
>> |
13 |
>> |
14 |
>> and 'xl info -n' shows two nodes: |
15 |
>> |
16 |
>> |
17 |
>> node: memsize memfree distances |
18 |
>> 0: 14656 3304 10,20 |
19 |
>> 1: 23424 7792 20,10 |
20 |
>> |
21 |
>> |
22 |
>> which could be right --- though I would expect each node to have 12GB |
23 |
>> rather than these weird sizes. |
24 |
>> |
25 |
>> So is NUMA turned off or not? Is it even possible to turn it off when |
26 |
>> it's |
27 |
>> enabled in the BIOS? |
28 |
> |
29 |
> If BIOS has it enabled. Then the OS still beeds to support it. |
30 |
> |
31 |
> Recent Xen has support for it. |
32 |
|
33 |
Is there any way to find out if it's actually made use of? |
34 |
|
35 |
> As for the non uniform spread of the memory. It depends how the memory modules are placed in the mainboard with regards to the Numa nodes. (CPU) |
36 |
|
37 |
The memory is evenly spread. The server has two CPUs with 6 memory banks each. |
38 |
All banks for each CPU are loaded identically. Hence two NUMA nodes would make |
39 |
sense, and each of them should see/have 12GB (now 14 each because I changed out |
40 |
the server because it kept crashing/freezing/becoming unreachable even after the |
41 |
software updates). |