| Message ID | 20220706085320.17581-1-atenart@kernel.org (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net-next] Documentation: add a description for net.core.high_order_alloc_disable | expand |
On Wed, Jul 6, 2022 at 10:53 AM Antoine Tenart <atenart@kernel.org> wrote: > > A description is missing for the net.core.high_order_alloc_disable > option in admin-guide/sysctl/net.rst ; add it. The above sysctl option > was introduced by commit ce27ec60648d ("net: add high_order_alloc_disable > sysctl/static key"). > > Cc: Eric Dumazet <edumazet@google.com> > Signed-off-by: Antoine Tenart <atenart@kernel.org> > --- > Documentation/admin-guide/sysctl/net.rst | 10 ++++++++++ > 1 file changed, 10 insertions(+) > > diff --git a/Documentation/admin-guide/sysctl/net.rst b/Documentation/admin-guide/sysctl/net.rst > index fcd650bdbc7e..85ab83411359 100644 > --- a/Documentation/admin-guide/sysctl/net.rst > +++ b/Documentation/admin-guide/sysctl/net.rst > @@ -391,6 +391,16 @@ GRO has decided not to coalesce, it is placed on a per-NAPI list. This > list is then passed to the stack when the number of segments reaches the > gro_normal_batch limit. > > +high_order_alloc_disable > +------------------------ > + > +By default the allocator for page frags tries to use high order pages (order-3 > +on x86). While the default behavior gives good results in most cases, some users > +might hit a contention in page allocations/freeing. This allows to opt-in for > +order-0 allocation instead. > Note: linux-5.14 allowed high-order pages to be stored on the per-cpu lists. I ran again the benchmark cited in commit ce27ec60648d to confirm that the slowdown we had before 5.14 for high number of alloc/frees per second is no more. for thr in {1..30} do sysctl -wq net.core.high_order_alloc_disable=0 T0=`./super_netperf $thr -H 127.0.0.1 -l 15` sysctl -wq net.core.high_order_alloc_disable=1 T1=`./super_netperf $thr -H 127.0.0.1 -l 15` echo $thr:$T0:$T1 done 1:62799:48115 2:121604:94588 3:183240:138376 4:241544:184639 5:302641:232985 6:358566:276494 7:418464:316401 8:470687:362562 9:528084:401193 10:586269:446422 11:642764:482556 12:696490:530701 13:748840:573388 14:808681:610787 15:856676:645813 16:912061:687695 17:966847:738943 18:1011960:771893 19:1056554:798815 20:1109519:844586 21:1159857:849445 22:1203224:888646 23:1226389:933538 24:1238174:964386 ... + > +Default: 0 > + > 2. /proc/sys/net/unix - Parameters for Unix domain sockets > ---------------------------------------------------------- > > -- > 2.36.1 >
On Wed, 6 Jul 2022 15:24:58 +0200 Eric Dumazet wrote: > linux-5.14 allowed high-order pages to be stored on the per-cpu lists. > > I ran again the benchmark cited in commit ce27ec60648d to confirm that > the slowdown we had before 5.14 for > high number of alloc/frees per second is no more. Sounds useful to know - Antoine, do you reckon we can include a mention of this knob being mostly of historical importance?
Quoting Jakub Kicinski (2022-07-06 23:18:39) > On Wed, 6 Jul 2022 15:24:58 +0200 Eric Dumazet wrote: > > linux-5.14 allowed high-order pages to be stored on the per-cpu lists. > > > > I ran again the benchmark cited in commit ce27ec60648d to confirm that > > the slowdown we had before 5.14 for > > high number of alloc/frees per second is no more. > > Sounds useful to know - Antoine, do you reckon we can include a mention > of this knob being mostly of historical importance? Sure, I'll send a v2 with that. Thanks! Antoine
diff --git a/Documentation/admin-guide/sysctl/net.rst b/Documentation/admin-guide/sysctl/net.rst index fcd650bdbc7e..85ab83411359 100644 --- a/Documentation/admin-guide/sysctl/net.rst +++ b/Documentation/admin-guide/sysctl/net.rst @@ -391,6 +391,16 @@ GRO has decided not to coalesce, it is placed on a per-NAPI list. This list is then passed to the stack when the number of segments reaches the gro_normal_batch limit. +high_order_alloc_disable +------------------------ + +By default the allocator for page frags tries to use high order pages (order-3 +on x86). While the default behavior gives good results in most cases, some users +might hit a contention in page allocations/freeing. This allows to opt-in for +order-0 allocation instead. + +Default: 0 + 2. /proc/sys/net/unix - Parameters for Unix domain sockets ----------------------------------------------------------
A description is missing for the net.core.high_order_alloc_disable option in admin-guide/sysctl/net.rst ; add it. The above sysctl option was introduced by commit ce27ec60648d ("net: add high_order_alloc_disable sysctl/static key"). Cc: Eric Dumazet <edumazet@google.com> Signed-off-by: Antoine Tenart <atenart@kernel.org> --- Documentation/admin-guide/sysctl/net.rst | 10 ++++++++++ 1 file changed, 10 insertions(+)