lkml.org 
[lkml]   [2021]   [Nov]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH v3] mm: migrate: Support multiple target nodes demotion
From


On 2021/11/12 10:44, Huang, Ying wrote:
> Baolin Wang <baolin.wang@linux.alibaba.com> writes:
>
>> We have some machines with multiple memory types like below, which
>> have one fast (DRAM) memory node and two slow (persistent memory) memory
>> nodes. According to current node demotion policy, if node 0 fills up,
>> its memory should be migrated to node 1, when node 1 fills up, its
>> memory will be migrated to node 2: node 0 -> node 1 -> node 2 ->stop.
>>
>> But this is not efficient and suitbale memory migration route
>> for our machine with multiple slow memory nodes. Since the distance
>> between node 0 to node 1 and node 0 to node 2 is equal, and memory
>> migration between slow memory nodes will increase persistent memory
>> bandwidth greatly, which will hurt the whole system's performance.
>>
>> Thus for this case, we can treat the slow memory node 1 and node 2
>> as a whole slow memory region, and we should migrate memory from
>> node 0 to node 1 and node 2 if node 0 fills up.
>>
>> This patch changes the node_demotion data structure to support multiple
>> target nodes, and establishes the migration path to support multiple
>> target nodes with validating if the node distance is the best or not.
>>
>> available: 3 nodes (0-2)
>> node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
>> node 0 size: 62153 MB
>> node 0 free: 55135 MB
>> node 1 cpus:
>> node 1 size: 127007 MB
>> node 1 free: 126930 MB
>> node 2 cpus:
>> node 2 size: 126968 MB
>> node 2 free: 126878 MB
>> node distances:
>> node 0 1 2
>> 0: 10 20 20
>> 1: 20 10 20
>> 2: 20 20 10
>>
>> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>

snip

>> /*
>> * 'next_pass' contains nodes which became migration
>> @@ -3192,6 +3281,14 @@ static int __init migrate_on_reclaim_init(void)
>> {
>> int ret;
>>
>> + /*
>> + * Ignore allocation failure, if this kmalloc fails
>> + * at boot time, we are likely in bigger trouble.
>> + */
>> + node_demotion = kmalloc_array(nr_node_ids,
>> + sizeof(struct demotion_nodes),
>> + GFP_KERNEL);
>> +
>
> I think we should WARN_ON() here.

In this unlikey case, I think the mm core will print more information,
IMHO WARN_ON() will help little. Anyway no strong opinion on this. Other
than that, can I get your reviewed-by tag with this nit fixed? Thanks.

\
 
 \ /
  Last update: 2021-11-12 03:58    [W:0.062 / U:0.300 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site