public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
* LLC cache miss increases with the increases of hugepages
@ 2017-03-06  2:53 linhaifeng
  0 siblings, 0 replies; only message in thread
From: linhaifeng @ 2017-03-06  2:53 UTC (permalink / raw)
  To: nyc, tglx, mingo, hpa; +Cc: linux-kernel

Hi,

I did a network forward test with ovs-dpdk, and found that LLC cache miss
increases with the increases of hugepages. ovs-dpdk only use 4 hugepages,
hugepage size is 1G.Why LLC cache miss increases ? I just echo more hugepages
but haven't been used yet.

some informations:

CPU: Intel(R) Xeon(R) CPU E5-2620 v3@ 2.40GHz
L1d cache:             32K
L1i cache:             32K
L2 cache:              256K
L3 cache:              15360K

echo 10 > /proc/sys/vm/nr_hugepages
ovs-dpdk's LLC-load-misses:3.88%
ovs-dpdk's iTLB-load-misses: 4.37%


echo 100 > /proc/sys/vm/nr_hugepages
reboot VM restart test.
ovs-dpdk's LLC-load-misses:31.55%
ovs-dpdk's iTLB-load-misses: 36.36%

I use perf to watch LLC-loads when not run test:

Samples: 1M of event 'LLC-loads', Event count (approx.): 57240206
Overhead  Shared Object                 Symbol
   9.19%  [kernel]                      [k] copy_page_rep
   5.03%  [kernel]                      [k] ktime_get
   4.06%  [unknown]                     [.] 0x000000000043e39d
   2.65%  bash                          [.] 0x000000000003e39d
   2.13%  [kernel]                      [k] rcu_check_callbacks

The function copy_page_rep seems like would prefetch all hugepages?

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2017-03-06  3:00 UTC | newest]

Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2017-03-06  2:53 LLC cache miss increases with the increases of hugepages linhaifeng

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox