Messages in this thread | | | Subject | Re: [PATCH] mm: migrate: Support multiple target nodes demotion | From | Baolin Wang <> | Date | Wed, 10 Nov 2021 18:45:17 +0800 |
| |
On 2021/11/10 16:51, Huang, Ying writes: > 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, if node 0 fills up, > ~~~~~~~~~~~~~ > > node demotion policy?
Yes, will fix in next version.
> > >> 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> >> --- >> Changes from RFC v2: >> - Change to 'short' type for target nodes array. >> - Remove nodemask instead selecting target node directly. >> - Add WARN_ONCE() if the target nodes exceed the maximum value. >> >> Changes from RFC v1: >> - Re-define the node_demotion structure. >> - Set up multiple target nodes by validating the node distance. >> - Add more comments. >> --- >> mm/migrate.c | 138 +++++++++++++++++++++++++++++++++++++++++++---------------- >> 1 file changed, 102 insertions(+), 36 deletions(-) >> >> diff --git a/mm/migrate.c b/mm/migrate.c >> index cf25b00..7f1d745 100644 >> --- a/mm/migrate.c >> +++ b/mm/migrate.c >> @@ -50,6 +50,7 @@ >> #include <linux/ptrace.h> >> #include <linux/oom.h> >> #include <linux/memory.h> >> +#include <linux/random.h> >> >> #include <asm/tlbflush.h> >> >> @@ -1119,12 +1120,25 @@ static int __unmap_and_move(struct page *page, struct page *newpage, >> * >> * This is represented in the node_demotion[] like this: >> * >> - * { 1, // Node 0 migrates to 1 >> - * 2, // Node 1 migrates to 2 >> - * -1, // Node 2 does not migrate >> - * 4, // Node 3 migrates to 4 >> - * 5, // Node 4 migrates to 5 >> - * -1} // Node 5 does not migrate >> + * { nr=1, nodes[0]=1 }, // Node 0 migrates to 1 >> + * { nr=1, nodes[0]=2 }, // Node 1 migrates to 2 >> + * { nr=0, nodes[0]=-1 }, // Node 2 does not migrate >> + * { nr=1, nodes[0]=4 }, // Node 3 migrates to 4 >> + * { nr=1, nodes[0]=5 }, // Node 4 migrates to 5 >> + * { nr=0, nodes[0]=-1} // Node 5 does not migrate >> + * >> + * Moreover some systems may have multiple same class memory >> + * types. Suppose a system has one socket with 3 memory nodes, > > s/same class memory types/slow memory nodes/ > > ? > > We don't support multiple fast memory types, right?
Until now we have no machines with multiple fast memory types. OK, I will change the words.
> >> + * node 0 is fast memory type, and node 1/2 both are slow memory >> + * type, and the distance between fast memory node and slow >> + * memory node is same. So the migration path should be: >> + * >> + * 0 -> 1/2 -> stop >> + * >> + * This is represented in the node_demotion[] like this: >> + * { nr=2, {nodes[0]=1, nodes[1]=2} }, // Node 0 migrates to node 1 and node 2 >> + * { nr=0, nodes[0]=-1, }, // Node 1 dose not migrate >> + * { nr=0, nodes[0]=-1, }, // Node 2 does not migrate >> */ >> >> /* >> @@ -1135,8 +1149,13 @@ static int __unmap_and_move(struct page *page, struct page *newpage, >> * must be held over all reads to ensure that no cycles are >> * observed. >> */ >> -static int node_demotion[MAX_NUMNODES] __read_mostly = >> - {[0 ... MAX_NUMNODES - 1] = NUMA_NO_NODE}; >> +#define DEMOTION_TARGET_NODES 15 >> +struct demotion_nodes { >> + unsigned short nr; >> + short nodes[DEMOTION_TARGET_NODES]; >> +}; >> + >> +static struct demotion_nodes node_demotion[MAX_NUMNODES] __read_mostly; > > If MAX_NUMNODES is 1024, the total size will be (16 * 2 * 1024) = 32K > bytes. That appears too large. We may consider to allocate > node_demotion[] dynamically.
Sure. I'd like to optimize it in a separate patch to keep current patch easy to review. Thanks.
| |