| Message ID | 1311876249.2346.39.camel@edumazet-HP-Compaq-6005-Pro-SFF-PC |
|---|---|
| State | Not Applicable, archived |
| Delegated to: | David Miller |
| Headers | show |
On Thu, Jul 28, 2011 at 08:04:09PM +0200, Eric Dumazet wrote: > Use NUMA aware allocations to reduce latencies and increase throughput. > > sunrpc kthreads can use kthread_create_on_node() if pool_mode is > "percpu" or "pernode", and svc_prepare_thread()/svc_init_buffer() can > also take into account NUMA node affinity for memory allocations. ... > @@ -662,14 +675,16 @@ svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) > nrservs--; > chosen_pool = choose_pool(serv, pool, &state); > > - rqstp = svc_prepare_thread(serv, chosen_pool); > + node = svc_pool_map_get_node(chosen_pool->sp_id); > + rqstp = svc_prepare_thread(serv, chosen_pool, node); The only correct value for the third argument there is svc_pool_map_get_node(chosen_pool->sp_id), so let's have svc_prepare_thread() call that itself. Seems OK otherwise. Any suggestions on how we should test this? --b. > if (IS_ERR(rqstp)) { > error = PTR_ERR(rqstp); > break; > } > > __module_get(serv->sv_module); > - task = kthread_create(serv->sv_function, rqstp, serv->sv_name); > + task = kthread_create_on_node(serv->sv_function, rqstp, > + node, serv->sv_name); > if (IS_ERR(task)) { > error = PTR_ERR(task); > module_put(serv->sv_module); > > -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Le vendredi 29 juillet 2011 à 12:42 -0400, J. Bruce Fields a écrit : > On Thu, Jul 28, 2011 at 08:04:09PM +0200, Eric Dumazet wrote: > > Use NUMA aware allocations to reduce latencies and increase throughput. > > > > sunrpc kthreads can use kthread_create_on_node() if pool_mode is > > "percpu" or "pernode", and svc_prepare_thread()/svc_init_buffer() can > > also take into account NUMA node affinity for memory allocations. > ... > > @@ -662,14 +675,16 @@ svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) > > nrservs--; > > chosen_pool = choose_pool(serv, pool, &state); > > > > - rqstp = svc_prepare_thread(serv, chosen_pool); > > + node = svc_pool_map_get_node(chosen_pool->sp_id); > > + rqstp = svc_prepare_thread(serv, chosen_pool, node); > > The only correct value for the third argument there is > svc_pool_map_get_node(chosen_pool->sp_id), so let's have > svc_prepare_thread() call that itself. > I have no idea of what you mean ;) I need 'node' for the following kthread_create_on_node() > Seems OK otherwise. > > Any suggestions on how we should test this? I did tests on my machine, seems good. I checked that stacks were now correct using : "echo t > /proc/sysrq-trigger" -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, Jul 29, 2011 at 08:02:05PM +0200, Eric Dumazet wrote: > Le vendredi 29 juillet 2011 à 12:42 -0400, J. Bruce Fields a écrit : > > On Thu, Jul 28, 2011 at 08:04:09PM +0200, Eric Dumazet wrote: > > > Use NUMA aware allocations to reduce latencies and increase throughput. > > > > > > sunrpc kthreads can use kthread_create_on_node() if pool_mode is > > > "percpu" or "pernode", and svc_prepare_thread()/svc_init_buffer() can > > > also take into account NUMA node affinity for memory allocations. > > ... > > > @@ -662,14 +675,16 @@ svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) > > > nrservs--; > > > chosen_pool = choose_pool(serv, pool, &state); > > > > > > - rqstp = svc_prepare_thread(serv, chosen_pool); > > > + node = svc_pool_map_get_node(chosen_pool->sp_id); > > > + rqstp = svc_prepare_thread(serv, chosen_pool, node); > > > > The only correct value for the third argument there is > > svc_pool_map_get_node(chosen_pool->sp_id), so let's have > > svc_prepare_thread() call that itself. > > > > I have no idea of what you mean ;) > > I need 'node' for the following kthread_create_on_node() Doh, of course--apologies. > > Seems OK otherwise. > > > > Any suggestions on how we should test this? > > I did tests on my machine, seems good. > > I checked that stacks were now correct using : > "echo t > /proc/sysrq-trigger" I was wondering more about good tests of nfsd's performance on numa; that might be more of a question for Greg. --b. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Sent from my iPhone On 30/07/2011, at 4:08, "J. Bruce Fields" <bfields@fieldses.org> wrote: > On Fri, Jul 29, 2011 at 08:02:05PM +0200, Eric Dumazet wrote: >> Le vendredi 29 juillet 2011 à 12:42 -0400, J. Bruce Fields a écri >> t : >>> On Thu, Jul 28, 2011 at 08:04:09PM +0200, Eric Dumazet wrote: >>>> Use NUMA aware allocations to reduce latencies and increase >>>> throughput. >>>> >>>> sunrpc kthreads can use kthread_create_on_node() if pool_mode is >>>> "percpu" or "pernode", and svc_prepare_thread()/svc_init_buffer() >>>> can >>>> also take into account NUMA node affinity for memory allocations. >>> ... >>>> @@ -662,14 +675,16 @@ svc_set_num_threads(struct svc_serv *serv, >>>> struct svc_pool *pool, int nrservs) >>>> nrservs--; >>>> chosen_pool = choose_pool(serv, pool, &state); >>>> >>>> - rqstp = svc_prepare_thread(serv, chosen_pool); >>>> + node = svc_pool_map_get_node(chosen_pool->sp_id); >>>> + rqstp = svc_prepare_thread(serv, chosen_pool, node); >>> >>> The only correct value for the third argument there is >>> svc_pool_map_get_node(chosen_pool->sp_id), so let's have >>> svc_prepare_thread() call that itself. >>> >> >> I have no idea of what you mean ;) >> >> I need 'node' for the following kthread_create_on_node() > > Doh, of course--apologies. > >>> Seems OK otherwise. >>> >>> Any suggestions on how we should test this? >> >> I did tests on my machine, seems good. >> >> I checked that stacks were now correct using : >> "echo t > /proc/sysrq-trigger" > > I was wondering more about good tests of nfsd's performance on numa; > that might be more of a question for Greg. > To really show a big difference you need a much bigger box, or slower NUMA interconnects than today's. You also want network cards locally attached to each node and a metadata heavy (i.e. high rpc call rate) load. Greg.-- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/fs/lockd/svc.c b/fs/lockd/svc.c index abfff9d..c061b9a 100644 --- a/fs/lockd/svc.c +++ b/fs/lockd/svc.c @@ -282,7 +282,7 @@ int lockd_up(void) /* * Create the kernel thread and wait for it to start. */ - nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0]); + nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0], NUMA_NO_NODE); if (IS_ERR(nlmsvc_rqst)) { error = PTR_ERR(nlmsvc_rqst); nlmsvc_rqst = NULL; diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c index e3d2942..ce620b5 100644 --- a/fs/nfs/callback.c +++ b/fs/nfs/callback.c @@ -125,7 +125,7 @@ nfs4_callback_up(struct svc_serv *serv) else goto out_err; - return svc_prepare_thread(serv, &serv->sv_pools[0]); + return svc_prepare_thread(serv, &serv->sv_pools[0], NUMA_NO_NODE); out_err: if (ret == 0) diff --git a/include/linux/sunrpc/svc.h b/include/linux/sunrpc/svc.h index 223588a..a78a51e 100644 --- a/include/linux/sunrpc/svc.h +++ b/include/linux/sunrpc/svc.h @@ -404,7 +404,7 @@ struct svc_procedure { struct svc_serv *svc_create(struct svc_program *, unsigned int, void (*shutdown)(struct svc_serv *)); struct svc_rqst *svc_prepare_thread(struct svc_serv *serv, - struct svc_pool *pool); + struct svc_pool *pool, int node); void svc_exit_thread(struct svc_rqst *); struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int, void (*shutdown)(struct svc_serv *), diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c index 6a69a11..30d70ab 100644 --- a/net/sunrpc/svc.c +++ b/net/sunrpc/svc.c @@ -295,6 +295,18 @@ svc_pool_map_put(void) } +static int svc_pool_map_get_node(unsigned int pidx) +{ + const struct svc_pool_map *m = &svc_pool_map; + + if (m->count) { + if (m->mode == SVC_POOL_PERCPU) + return cpu_to_node(m->pool_to[pidx]); + if (m->mode == SVC_POOL_PERNODE) + return m->pool_to[pidx]; + } + return NUMA_NO_NODE; +} /* * Set the given thread's cpus_allowed mask so that it * will only run on cpus in the given pool. @@ -499,7 +511,7 @@ EXPORT_SYMBOL_GPL(svc_destroy); * We allocate pages and place them in rq_argpages. */ static int -svc_init_buffer(struct svc_rqst *rqstp, unsigned int size) +svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node) { unsigned int pages, arghi; @@ -513,7 +525,7 @@ svc_init_buffer(struct svc_rqst *rqstp, unsigned int size) arghi = 0; BUG_ON(pages > RPCSVC_MAXPAGES); while (pages) { - struct page *p = alloc_page(GFP_KERNEL); + struct page *p = alloc_pages_node(node, GFP_KERNEL, 0); if (!p) break; rqstp->rq_pages[arghi++] = p; @@ -536,11 +548,11 @@ svc_release_buffer(struct svc_rqst *rqstp) } struct svc_rqst * -svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool) +svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node) { struct svc_rqst *rqstp; - rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL); + rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node); if (!rqstp) goto out_enomem; @@ -554,15 +566,15 @@ svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool) rqstp->rq_server = serv; rqstp->rq_pool = pool; - rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL); + rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); if (!rqstp->rq_argp) goto out_thread; - rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL); + rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); if (!rqstp->rq_resp) goto out_thread; - if (!svc_init_buffer(rqstp, serv->sv_max_mesg)) + if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node)) goto out_thread; return rqstp; @@ -647,6 +659,7 @@ svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) struct svc_pool *chosen_pool; int error = 0; unsigned int state = serv->sv_nrthreads-1; + int node; if (pool == NULL) { /* The -1 assumes caller has done a svc_get() */ @@ -662,14 +675,16 @@ svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) nrservs--; chosen_pool = choose_pool(serv, pool, &state); - rqstp = svc_prepare_thread(serv, chosen_pool); + node = svc_pool_map_get_node(chosen_pool->sp_id); + rqstp = svc_prepare_thread(serv, chosen_pool, node); if (IS_ERR(rqstp)) { error = PTR_ERR(rqstp); break; } __module_get(serv->sv_module); - task = kthread_create(serv->sv_function, rqstp, serv->sv_name); + task = kthread_create_on_node(serv->sv_function, rqstp, + node, serv->sv_name); if (IS_ERR(task)) { error = PTR_ERR(task); module_put(serv->sv_module);
Use NUMA aware allocations to reduce latencies and increase throughput. sunrpc kthreads can use kthread_create_on_node() if pool_mode is "percpu" or "pernode", and svc_prepare_thread()/svc_init_buffer() can also take into account NUMA node affinity for memory allocations. Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> CC: "J. Bruce Fields" <bfields@fieldses.org> CC: Neil Brown <neilb@suse.de> CC: David Miller <davem@davemloft.net> --- fs/lockd/svc.c | 2 +- fs/nfs/callback.c | 2 +- include/linux/sunrpc/svc.h | 2 +- net/sunrpc/svc.c | 33 ++++++++++++++++++++++++--------- 4 files changed, 27 insertions(+), 12 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html