From: James Simmons <jsimmons@infradead.org>
To: Andreas Dilger <adilger@whamcloud.com>,
Oleg Drokin <green@whamcloud.com>, NeilBrown <neilb@suse.de>
Cc: Serguei Smirnov <ssmirnov@whamcloud.com>,
Lustre Development List <lustre-devel@lists.lustre.org>
Subject: [lustre-devel] [PATCH 14/15] lnet: socklnd: set conns_per_peer based on link speed
Date: Sun, 22 Aug 2021 22:27:45 -0400 [thread overview]
Message-ID: <1629685666-4533-15-git-send-email-jsimmons@infradead.org> (raw)
In-Reply-To: <1629685666-4533-1-git-send-email-jsimmons@infradead.org>
From: Serguei Smirnov <ssmirnov@whamcloud.com>
Specifying conns_per_peer=0 for a ni is now used to set
the conns_per_peer as a function of the corresponding link speed
as follows:
conns_per_peer = (ilog2(Gbps) / 2 + 1)
Listed below are the resulting defaults for common link speeds:
100Gbps, 200Gbps -> 4
50Gbps -> 3
5Gbps, 10Gbps -> 2
less than 4Gbps -> 1
WC-bug-id: https://jira.whamcloud.com/browse/LU-12815
Lustre-commit: c44afcfb72a1c2fd ("LU-12815 socklnd: set conns_per_peer based on link speed")
Signed-off-by: Serguei Smirnov <ssmirnov@whamcloud.com>
Reviewed-on: https://review.whamcloud.com/44417
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: James Simmons <jsimmons@infradead.org>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
Signed-off-by: James Simmons <jsimmons@infradead.org>
---
net/lnet/klnds/socklnd/socklnd_modparams.c | 75 +++++++++++++++++++++++++++++-
1 file changed, 73 insertions(+), 2 deletions(-)
diff --git a/net/lnet/klnds/socklnd/socklnd_modparams.c b/net/lnet/klnds/socklnd/socklnd_modparams.c
index c6cce1e..72f9df2 100644
--- a/net/lnet/klnds/socklnd/socklnd_modparams.c
+++ b/net/lnet/klnds/socklnd/socklnd_modparams.c
@@ -23,6 +23,8 @@
#if defined(__x86_64__) || defined(__i386__)
#include <asm/hypervisor.h>
#endif
+#include <linux/inetdevice.h>
+#include <linux/ethtool.h>
#define CURRENT_LND_VERSION 1
@@ -154,6 +156,75 @@
struct ksock_tunables ksocknal_tunables;
static struct lnet_ioctl_config_socklnd_tunables default_tunables;
+static int ksocklnd_ni_get_eth_intf_speed(struct lnet_ni *ni)
+{
+ const struct in_ifaddr *ifa;
+ struct net_device *dev;
+ int intf_idx = -1;
+ int ret = -1;
+
+ rtnl_lock();
+ for_each_netdev(ni->ni_net_ns, dev) {
+ int flags = dev_get_flags(dev);
+ struct in_device *in_dev;
+
+ if (flags & IFF_LOOPBACK) /* skip the loopback IF */
+ continue;
+
+ if (!(flags & IFF_UP))
+ continue;
+
+ in_dev = __in_dev_get_rcu(dev);
+ if (!in_dev)
+ continue;
+
+ in_dev_for_each_ifa_rcu(ifa, in_dev) {
+ if (strcmp(ifa->ifa_label, ni->ni_interface) == 0)
+ intf_idx = dev->ifindex;
+ }
+ if (intf_idx >= 0)
+ break;
+ }
+ if (intf_idx >= 0) {
+ struct ethtool_link_ksettings cmd;
+ int ethtool_ret;
+
+ /* Some devices may not be providing link settings */
+ ethtool_ret = __ethtool_get_link_ksettings(dev, &cmd);
+ if (!ethtool_ret)
+ ret = cmd.base.speed;
+ else
+ ret = ethtool_ret;
+ }
+ rtnl_unlock();
+
+ return ret;
+}
+
+static int ksocklnd_speed2cpp(int speed)
+{
+ /* Use the minimum of 1Gbps to avoid calling ilog2 with 0 */
+ if (speed < 1000)
+ speed = 1000;
+
+ /* Pick heuristically optimal conns_per_peer value
+ * for the specified ethernet interface speed (Mbps)
+ */
+ return ilog2(speed / 1000) / 2 + 1;
+}
+
+static int ksocklnd_lookup_conns_per_peer(struct lnet_ni *ni)
+{
+ int cpp = DEFAULT_CONNS_PER_PEER;
+ int speed = ksocklnd_ni_get_eth_intf_speed(ni);
+
+ CDEBUG(D_NET, "intf %s speed %d\n", ni->ni_interface, speed);
+ if (speed > 0)
+ cpp = ksocklnd_speed2cpp(speed);
+
+ return cpp;
+}
+
int ksocknal_tunables_init(void)
{
default_tunables.lnd_version = CURRENT_LND_VERSION;
@@ -248,6 +319,6 @@ void ksocknal_tunables_setup(struct lnet_ni *ni)
*ksocknal_tunables.ksnd_peerrtrcredits;
if (!tunables->lnd_conns_per_peer)
- tunables->lnd_conns_per_peer = (conns_per_peer) ?
- conns_per_peer : DEFAULT_CONNS_PER_PEER;
+ tunables->lnd_conns_per_peer =
+ ksocklnd_lookup_conns_per_peer(ni);
}
--
1.8.3.1
_______________________________________________
lustre-devel mailing list
lustre-devel@lists.lustre.org
http://lists.lustre.org/listinfo.cgi/lustre-devel-lustre.org
next prev parent reply other threads:[~2021-08-23 2:28 UTC|newest]
Thread overview: 16+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-08-23 2:27 [lustre-devel] [PATCH 00/15] lustre: sync to OpenSFS as of Aug 22, 2021 James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 01/15] lustre: uapi: support fixed directory layout James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 02/15] lustre: pcc: add LCM_FL_PCC_RDONLY layout flag James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 03/15] lustre: mdt: implement fallocate in MDC/MDT James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 04/15] lnet: Reflect ni_fatal in NI status James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 05/15] lustre: obdclass: reintroduce lu_ref James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 06/15] lnet: keep in insync to change due to GPU Direct Support James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 07/15] lustre: osc: Support RDMA only pages James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 08/15] lustre: mgc: rework mgc_apply_recover_logs() for gcc10 James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 09/15] lnet: socklnd: allow dynamic setting of conns_per_peer James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 10/15] lnet: Provide kernel API for adding peers James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 11/15] lustre: obdclass: Add peer/peer NI when processing llog James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 12/15] lnet: peer state to lock primary nid James Simmons
2021-08-23 2:27 ` [lustre-devel] [PATCH 13/15] lustre: llite: Proved an abstraction for AS_EXITING James Simmons
2021-08-23 2:27 ` James Simmons [this message]
2021-08-23 2:27 ` [lustre-devel] [PATCH 15/15] lustre: update version to 2.14.54 James Simmons
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1629685666-4533-15-git-send-email-jsimmons@infradead.org \
--to=jsimmons@infradead.org \
--cc=adilger@whamcloud.com \
--cc=green@whamcloud.com \
--cc=lustre-devel@lists.lustre.org \
--cc=neilb@suse.de \
--cc=ssmirnov@whamcloud.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).