On 10/9/24 18:18, Alexander Atanasov wrote:
On 9.10.24 19:09, Denis V. Lunev wrote:
On 10/7/24 13:12, Alexander Atanasov wrote:
Convert to lockless lists - intermix with regular list due to
that next pointer in both list_head and llist_head is the first
field, and prev is not used. Do this so we can make babysteps
forward.
[cut]
I observe
[w(1)][52.5%][w=644MiB/s][w=165k IOPS][eta 00m:28s] <- 1 thread,
partition
1 (f=1): [w(1)][15.0%][w=715MiB/s][w=183k IOPS][eta 00m:51s] <-
thread device
[w(12)][25.0%][w=1376MiB/s][w=352k IOPS][eta 00m:45s] <- 12 threads
Do we expect to have a difference right now?
These are minor and preparations - converted lists are about metadata
writeback only.
I expect improvement when the use of deferred lock is
eliminated/reduced to minimum which will happen when the rest of the
lists are converted, i.e. when all pios are moved lockless and
contention is reduced.
but I expect at least marginal improvement in single
thread once we have stable mappings, which has not been
happened.
Den
_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel
