From: David Christensen Date: Wed, 28 Apr 2021 20:10:16 +0000 (-0700) Subject: config/ppc: reduce number of cores and NUMA nodes X-Git-Url: http://git.droids-corp.org/?a=commitdiff_plain;h=5f760b7dccd85c7da78c6e4ee3517a912861d424;p=dpdk.git config/ppc: reduce number of cores and NUMA nodes When setting RTE_MAX_LCORES to the maximum value supported by ppc hardware (1536), the lcores_autotest may timeout after 30 seconds because the test takes nearly 60 seconds to complete. Set max_lcores to a lower value because the maximum value is unlikely to be seen in any production systems and to eliminate the quick test timeout error. Bugzilla ID: 684 Fixes: db1f2f8a9fe5 ("config: increase maximum lcores for ppc") Cc: stable@dpdk.org Signed-off-by: David Christensen Acked-by: Luca Boccassi --- diff --git a/config/ppc/meson.build b/config/ppc/meson.build index 1c196b8db5..4f7806bab1 100644 --- a/config/ppc/meson.build +++ b/config/ppc/meson.build @@ -17,7 +17,18 @@ if not power9_supported dpdk_conf.set('RTE_MACHINE','power8') endif -# overrides specific to ppc64 -dpdk_conf.set('RTE_MAX_LCORE', 1536) -dpdk_conf.set('RTE_MAX_NUMA_NODES', 32) +# Certain POWER9 systems can scale as high as 1536 LCORES, but setting such a +# high value can waste memory, cause timeouts in time limited autotests, and is +# unlikely to be used in many production situations. Similarly, keeping the +# default 64 LCORES seems too small as most POWER9 dual socket systems will have +# at least 128 LCORES available. Set RTE_MAX_LCORE to 128 for POWER systems as +# a compromise. +dpdk_conf.set('RTE_MAX_LCORE', 128) + +# POWER systems do not allocate NUMA nodes sequentially. A dual socket system +# will have CPUs associated with NUMA nodes 0 & 8, so ensure that the second +# NUMA node will be supported by setting RTE_MAX_NUMA_NODES to 16. High end +# systems can scale even higher with as many as 32 NUMA nodes. +dpdk_conf.set('RTE_MAX_NUMA_NODES', 16) + dpdk_conf.set('RTE_CACHE_LINE_SIZE', 128)