All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH 000 of 11] knfsd: Introduction - Make knfsd more NUMA-aware
@ 2006-07-31  0:41 ` NeilBrown
  0 siblings, 0 replies; 51+ messages in thread
From: NeilBrown @ 2006-07-31  0:41 UTC (permalink / raw)
  To: Andrew Morton; +Cc: nfs, linux-kernel

Following are 11 patches from Greg Banks which combine to make knfsd
more Numa-aware.  They reduce hitting on 'global' data structures, and
create some data-structures that can be node-local.

knfsd threads are bound to a particular node, and the thread to handle
a new request is chosen from the threads that are attach to the node
that received the interrupt.

The distribution of threads across nodes can be controlled by a new
file in the 'nfsd' filesystem, though the default approach of an even
spread is probably fine for most sites.

Some (old) numbers that show the efficacy of these patches:
N == number of NICs == number of CPUs == nmber of clients.
Number of NUMA nodes == N/2

N	Throughput, MiB/s	CPU usage, % (max=N*100)
	Before	After		Before	After
---	------	----		-----	-----
4	312	435		350	228
6	500	656		501	418
8	562	804		690	589


 [PATCH 001 of 11] knfsd: move tempsock aging to a timer
 [PATCH 002 of 11] knfsd: convert sk_inuse to atomic_t
 [PATCH 003 of 11] knfsd: use new lock for svc_sock deferred list
 [PATCH 004 of 11] knfsd: convert sk_reserved to atomic_t
 [PATCH 005 of 11] knfsd: test and set SK_BUSY atomically
 [PATCH 006 of 11] knfsd: split svc_serv into pools
 [PATCH 007 of 11] knfsd: add svc_get
 [PATCH 008 of 11] knfsd: add svc_set_num_threads
 [PATCH 009 of 11] knfsd: use svc_set_num_threads to manage threads in knfsd
 [PATCH 010 of 11] knfsd: make rpc threads pools numa aware
 [PATCH 011 of 11] knfsd: allow admin to set nthreads per node

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

^ permalink raw reply	[flat|nested] 51+ messages in thread
* [PATCH 008 of 11] knfsd: add svc_set_num_threads
@ 2006-07-25  5:14 Greg Banks
  0 siblings, 0 replies; 51+ messages in thread
From: Greg Banks @ 2006-07-25  5:14 UTC (permalink / raw)
  To: Neil Brown; +Cc: Linux NFS Mailing List

knfsd:  Currently knfsd keeps its own list of all nfsd threads
in nfssvc.c; add a new way of managing the list of all threads
in a svc_serv.  Add svc_create_pooled() to allow creation of
a svc_serv whose threads are managed by the sunrpc code.  Add
svc_set_num_threads() to manage the number of threads in a service,
either per-pool or globally across the service.


Signed-off-by: Greg Banks <gnb@melbourne.sgi.com>
---

 include/linux/sunrpc/svc.h |   19 +++-
 net/sunrpc/sunrpc_syms.c   |    2 
 net/sunrpc/svc.c           |  137 +++++++++++++++++++++++++++++++++-
 3 files changed, 150 insertions(+), 8 deletions(-)

Index: linus-git/include/linux/sunrpc/svc.h
===================================================================
--- linus-git.orig/include/linux/sunrpc/svc.h	2006-07-24 20:54:04.638635111 +1000
+++ linus-git/include/linux/sunrpc/svc.h	2006-07-24 22:16:36.041218126 +1000
@@ -17,6 +17,10 @@
 #include <linux/wait.h>
 #include <linux/mm.h>
 
+/*
+ * This is the RPC server thread function prototype
+ */
+typedef void		(*svc_thread_fn)(struct svc_rqst *);
 
 /*
  *
@@ -34,6 +38,7 @@ struct svc_pool {
 	struct list_head	sp_threads;	/* idle server threads */
 	struct list_head	sp_sockets;	/* pending sockets */
 	unsigned int		sp_nrthreads;	/* # of threads in pool */
+	struct list_head	sp_all_threads;	/* all server threads */
 } ____cacheline_aligned_in_smp;
 
 /*
@@ -62,6 +67,10 @@ struct svc_serv {
 	char *			sv_name;	/* service name */
 	unsigned int		sv_nrpools;	/* number of thread pools */
 	struct svc_pool *	sv_pools;	/* array of thread pools */
+	struct module *		sv_module;	/* optional module to count when
+						 * adding threads */
+	svc_thread_fn		sv_function;	/* main function for threads */
+	int			sv_kill_signal;	/* signal to kill threads */
 };
 
 /*
@@ -141,6 +150,7 @@ static inline void svc_putu32(struct kve
  */
 struct svc_rqst {
 	struct list_head	rq_list;	/* idle list */
+	struct list_head	rq_all;		/* all threads list */
 	struct svc_sock *	rq_sock;	/* socket */
 	struct sockaddr_in	rq_addr;	/* peer address */
 	int			rq_addrlen;
@@ -195,6 +205,7 @@ struct svc_rqst {
 						 * to prevent encrypting page
 						 * cache pages */
 	wait_queue_head_t	rq_wait;	/* synchronization */
+	struct task_struct	*rq_task;	/* service thread */
 };
 
 /*
@@ -336,16 +347,14 @@ struct svc_procedure {
 };
 
 /*
- * This is the RPC server thread function prototype
- */
-typedef void		(*svc_thread_fn)(struct svc_rqst *);
-
-/*
  * Function prototypes.
  */
 struct svc_serv *  svc_create(struct svc_program *, unsigned int);
 int		   svc_create_thread(svc_thread_fn, struct svc_serv *);
 void		   svc_exit_thread(struct svc_rqst *);
+struct svc_serv *  svc_create_pooled(struct svc_program *, unsigned int,
+			svc_thread_fn, int sig, struct module *);
+int		   svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
 void		   svc_destroy(struct svc_serv *);
 int		   svc_process(struct svc_serv *, struct svc_rqst *);
 int		   svc_register(struct svc_serv *, int, unsigned short);
Index: linus-git/net/sunrpc/sunrpc_syms.c
===================================================================
--- linus-git.orig/net/sunrpc/sunrpc_syms.c	2006-07-24 20:43:35.596757996 +1000
+++ linus-git/net/sunrpc/sunrpc_syms.c	2006-07-24 22:16:36.153203582 +1000
@@ -73,6 +73,8 @@ EXPORT_SYMBOL(put_rpccred);
 /* RPC server stuff */
 EXPORT_SYMBOL(svc_create);
 EXPORT_SYMBOL(svc_create_thread);
+EXPORT_SYMBOL(svc_create_pooled);
+EXPORT_SYMBOL(svc_set_num_threads);
 EXPORT_SYMBOL(svc_exit_thread);
 EXPORT_SYMBOL(svc_destroy);
 EXPORT_SYMBOL(svc_drop);
Index: linus-git/net/sunrpc/svc.c
===================================================================
--- linus-git.orig/net/sunrpc/svc.c	2006-07-24 20:54:04.874604317 +1000
+++ linus-git/net/sunrpc/svc.c	2006-07-24 22:16:36.157203063 +1000
@@ -12,6 +12,8 @@
 #include <linux/net.h>
 #include <linux/in.h>
 #include <linux/mm.h>
+#include <linux/interrupt.h>
+#include <linux/module.h>
 
 #include <linux/sunrpc/types.h>
 #include <linux/sunrpc/xdr.h>
@@ -25,8 +27,8 @@
 /*
  * Create an RPC service
  */
-struct svc_serv *
-svc_create(struct svc_program *prog, unsigned int bufsize)
+static struct svc_serv *
+__svc_create(struct svc_program *prog, unsigned int bufsize, int npools)
 {
 	struct svc_serv	*serv;
 	int vers;
@@ -59,7 +61,7 @@ svc_create(struct svc_program *prog, uns
 	init_timer(&serv->sv_temptimer);
 	spin_lock_init(&serv->sv_lock);
 
-	serv->sv_nrpools = 1;
+	serv->sv_nrpools = npools;
 	if (!(serv->sv_pools = (struct svc_pool *)
 		kmalloc(sizeof(struct svc_pool) * serv->sv_nrpools,
 			GFP_KERNEL))) {
@@ -77,6 +79,7 @@ svc_create(struct svc_program *prog, uns
 		pool->sp_id = i;
 		INIT_LIST_HEAD(&pool->sp_threads);
 		INIT_LIST_HEAD(&pool->sp_sockets);
+		INIT_LIST_HEAD(&pool->sp_all_threads);
 		spin_lock_init(&pool->sp_lock);
 	}
 
@@ -87,6 +90,29 @@ svc_create(struct svc_program *prog, uns
 	return serv;
 }
 
+struct svc_serv *
+svc_create(struct svc_program *prog, unsigned int bufsize)
+{
+	return __svc_create(prog, bufsize, /*npools*/1);
+}
+
+struct svc_serv *
+svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
+		  svc_thread_fn func, int sig, struct module *mod)
+{
+	struct svc_serv *serv;
+
+	serv = __svc_create(prog, bufsize, /*npools*/1);
+
+	if (serv != NULL) {
+		serv->sv_function = func;
+		serv->sv_kill_signal = sig;
+		serv->sv_module = mod;
+	}
+
+	return serv;
+}
+
 /*
  * Destroy an RPC service.  Should be called with the BKL held
  */
@@ -197,6 +223,7 @@ __svc_create_thread(svc_thread_fn func, 
 	serv->sv_nrthreads++;
 	spin_lock_bh(&pool->sp_lock);
 	pool->sp_nrthreads++;
+	list_add(&rqstp->rq_all, &pool->sp_all_threads);
 	spin_unlock_bh(&pool->sp_lock);
 	rqstp->rq_server = serv;
 	rqstp->rq_pool = pool;
@@ -223,6 +250,109 @@ svc_create_thread(svc_thread_fn func, st
 }
 
 /*
+ * Choose a pool in which to create a new thread, for svc_set_num_threads
+ */
+static inline struct svc_pool *
+choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
+{
+	if (pool != NULL)
+		return pool;
+
+ 	return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
+}
+
+/*
+ * Choose a thread to kill, for svc_set_num_threads
+ */
+static inline struct task_struct *
+choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
+{
+	unsigned int i;
+	struct task_struct *task = NULL;
+
+	if (pool != NULL) {
+		spin_lock_bh(&pool->sp_lock);
+	} else {
+		/* choose a pool in round-robin fashion */
+ 		for (i = 0 ; i < serv->sv_nrpools ; i++) {
+ 			pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
+			spin_lock_bh(&pool->sp_lock);
+ 			if (!list_empty(&pool->sp_all_threads))
+ 				goto found_pool;
+			spin_unlock_bh(&pool->sp_lock);
+ 		}
+		return NULL;
+	}
+
+found_pool:
+	if (!list_empty(&pool->sp_all_threads)) {
+		struct svc_rqst *rqstp;
+
+		/*
+		 * Remove from the pool->sp_all_threads list
+		 * so we don't try to kill it again.
+		 */
+		rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
+		list_del_init(&rqstp->rq_all);
+		task = rqstp->rq_task;
+    	}
+	spin_unlock_bh(&pool->sp_lock);
+
+	return task;
+}
+
+/*
+ * Create or destroy enough new threads to make the number
+ * of threads the given number.  If `pool' is non-NULL, applies
+ * only to threads in that pool, otherwise round-robins between
+ * all pools.  Must be called with a svc_get() reference and
+ * the BKL held.
+ *
+ * Destroying threads relies on the service threads filling in
+ * rqstp->rq_task, which only the nfs ones do.  Assumes the serv
+ * has been created using svc_create_pooled().
+ *
+ * Based on code that used to be in nfsd_svc() but tweaked
+ * to be pool-aware.
+ */
+int
+svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
+{
+	struct task_struct *victim;
+	int error = 0;
+	unsigned int state = serv->sv_nrthreads-1;
+
+	if (pool == NULL) {
+		/* The -1 assumes caller has done a svc_get() */
+		nrservs -= (serv->sv_nrthreads-1);
+	} else {
+		spin_lock_bh(&pool->sp_lock);
+		nrservs -= pool->sp_nrthreads;
+		spin_unlock_bh(&pool->sp_lock);
+	}
+
+	/* create new threads */
+	while (nrservs > 0) {
+		nrservs--;
+		__module_get(serv->sv_module);
+		error = __svc_create_thread(serv->sv_function, serv,
+					    choose_pool(serv, pool, &state));
+		if (error < 0) {
+			module_put(serv->sv_module);
+			break;
+		}
+	}
+	/* destroy old threads */
+	while (nrservs < 0 &&
+	       (victim = choose_victim(serv, pool, &state)) != NULL) {
+		send_sig(serv->sv_kill_signal, victim, 1);
+		nrservs++;
+	}
+
+	return error;
+}
+
+/*
  * Called from a server thread as it's exiting.  Caller must hold BKL.
  */
 void
@@ -238,6 +368,7 @@ svc_exit_thread(struct svc_rqst *rqstp)
 
 	spin_lock_bh(&pool->sp_lock);
 	pool->sp_nrthreads--;
+	list_del(&rqstp->rq_all);
 	spin_unlock_bh(&pool->sp_lock);
 
 	kfree(rqstp);

-- 
Greg Banks, R&D Software Engineer, SGI Australian Software Group.
I don't speak for SGI.



-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

^ permalink raw reply	[flat|nested] 51+ messages in thread
end of thread, other threads:[~2006-08-07 11:25 UTC | newest]

Thread overview: 51+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2006-07-31  0:41 [PATCH 000 of 11] knfsd: Introduction - Make knfsd more NUMA-aware NeilBrown
2006-07-31  0:41 ` NeilBrown
2006-07-31  0:41 ` [PATCH 001 of 11] knfsd: move tempsock aging to a timer NeilBrown
2006-07-31  0:41   ` NeilBrown
2006-07-31  0:41 ` [PATCH 002 of 11] knfsd: convert sk_inuse to atomic_t NeilBrown
2006-07-31  0:41   ` NeilBrown
2006-07-31  0:41 ` [PATCH 003 of 11] knfsd: use new lock for svc_sock deferred list NeilBrown
2006-07-31  0:41   ` NeilBrown
2006-07-31  0:42 ` [PATCH 004 of 11] knfsd: convert sk_reserved to atomic_t NeilBrown
2006-07-31  0:42   ` NeilBrown
2006-07-31  0:42 ` [PATCH 005 of 11] knfsd: test and set SK_BUSY atomically NeilBrown
2006-07-31  0:42   ` NeilBrown
2006-07-31  0:42 ` [PATCH 006 of 11] knfsd: split svc_serv into pools NeilBrown
2006-07-31  0:42   ` NeilBrown
2006-07-31  0:42 ` [PATCH 007 of 11] knfsd: add svc_get NeilBrown
2006-07-31  0:42   ` NeilBrown
2006-07-31  4:05   ` Andrew Morton
2006-07-31  4:05     ` Andrew Morton
2006-07-31  4:16     ` Neil Brown
2006-07-31  4:16       ` Neil Brown
2006-07-31  0:42 ` [PATCH 008 of 11] knfsd: add svc_set_num_threads NeilBrown
2006-07-31  0:42   ` NeilBrown
2006-07-31  4:11   ` Andrew Morton
2006-07-31  4:11     ` Andrew Morton
2006-07-31  4:24     ` Neil Brown
2006-07-31  4:24       ` [NFS] " Neil Brown
2006-07-31  0:42 ` [PATCH 009 of 11] knfsd: use svc_set_num_threads to manage threads in knfsd NeilBrown
2006-07-31  0:42   ` NeilBrown
2006-07-31  0:42 ` [PATCH 010 of 11] knfsd: make rpc threads pools numa aware NeilBrown
2006-07-31  0:42   ` NeilBrown
2006-07-31  4:14   ` Andrew Morton
2006-07-31  4:14     ` Andrew Morton
2006-07-31  4:36     ` Neil Brown
2006-07-31  4:36       ` Neil Brown
2006-07-31  4:42       ` Greg Banks
2006-07-31  4:42         ` [NFS] " Greg Banks
2006-07-31  5:54         ` Greg Banks
2006-07-31  5:54           ` [NFS] " Greg Banks
2006-08-01  4:43           ` Andrew Morton
2006-08-01  4:43             ` [NFS] " Andrew Morton
2006-08-01  5:22             ` Greg Banks
2006-08-01  5:22               ` [NFS] " Greg Banks
2006-08-06  9:47   ` Andrew Morton
2006-08-06  9:47     ` Andrew Morton
2006-08-07  3:16     ` Greg Banks
2006-08-07  3:16       ` Greg Banks
2006-08-07 11:25     ` Greg Banks
2006-08-07 11:25       ` Greg Banks
2006-07-31  0:42 ` [PATCH 011 of 11] knfsd: allow admin to set nthreads per node NeilBrown
2006-07-31  0:42   ` NeilBrown
  -- strict thread matches above, loose matches on Subject: below --
2006-07-25  5:14 [PATCH 008 of 11] knfsd: add svc_set_num_threads Greg Banks

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.