Re: IO scheduler based IO Controller V2

From: Andrea Righi <righi.andrea@gmail.com>
To: Vivek Goyal <vgoyal@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	nauman@google.com, dpshah@google.com, lizf@cn.fujitsu.com,
	mikew@google.com, fchecconi@gmail.com, paolo.valente@unimore.it,
	jens.axboe@oracle.com, ryov@valinux.co.jp,
	fernando@oss.ntt.co.jp, s-uchida@ap.jp.nec.com,
	taka@valinux.co.jp, guijianfeng@cn.fujitsu.com,
	jmoyer@redhat.com, dhaval@linux.vnet.ibm.com,
	balbir@linux.vnet.ibm.com, linux-kernel@vger.kernel.org,
	containers@lists.linux-foundation.org, agk@redhat.com,
	dm-devel@redhat.com, snitzer@redhat.com, m-ikeda@ds.jp.nec.com,
	peterz@infradead.org
Subject: Re: IO scheduler based IO Controller V2
Date: Thu, 7 May 2009 14:22:55 +0200	[thread overview]
Message-ID: <20090507122254.GA5892@linux> (raw)
In-Reply-To: <20090507090450.GA4613@linux>

On Thu, May 07, 2009 at 11:04:50AM +0200, Andrea Righi wrote:
> On Wed, May 06, 2009 at 05:52:35PM -0400, Vivek Goyal wrote:
> > > > Without io-throttle patches
> > > > ---------------------------
> > > > - Two readers, first BE prio 7, second BE prio 0
> > > > 
> > > > 234179072 bytes (234 MB) copied, 4.12074 s, 56.8 MB/s
> > > > High prio reader finished
> > > > 234179072 bytes (234 MB) copied, 5.36023 s, 43.7 MB/s
> > > > 
> > > > Note: There is no service differentiation between prio 0 and prio 7 task
> > > >       with io-throttle patches.
> > > > 
> > > > Test 3
> > > > ======
> > > > - Run the one RT reader and one BE reader in root cgroup without any
> > > >   limitations. I guess this should mean unlimited BW and behavior should
> > > >   be same as with CFQ without io-throttling patches.
> > > > 
> > > > With io-throttle patches
> > > > =========================
> > > > Ran the test 4 times because I was getting different results in different
> > > > runs.
> > > > 
> > > > - Two readers, one RT prio 0  other BE prio 7
> > > > 
> > > > 234179072 bytes (234 MB) copied, 2.74604 s, 85.3 MB/s
> > > > 234179072 bytes (234 MB) copied, 5.20995 s, 44.9 MB/s
> > > > RT task finished
> > > > 
> > > > 234179072 bytes (234 MB) copied, 4.54417 s, 51.5 MB/s
> > > > RT task finished
> > > > 234179072 bytes (234 MB) copied, 5.23396 s, 44.7 MB/s
> > > > 
> > > > 234179072 bytes (234 MB) copied, 5.17727 s, 45.2 MB/s
> > > > RT task finished
> > > > 234179072 bytes (234 MB) copied, 5.25894 s, 44.5 MB/s
> > > > 
> > > > 234179072 bytes (234 MB) copied, 2.74141 s, 85.4 MB/s
> > > > 234179072 bytes (234 MB) copied, 5.20536 s, 45.0 MB/s
> > > > RT task finished
> > > > 
> > > > Note: Out of 4 runs, looks like twice it is complete priority inversion
> > > >       and RT task finished after BE task. Rest of the two times, the
> > > >       difference between BW of RT and BE task is much less as compared to
> > > >       without patches. In fact once it was almost same.
> > > 
> > > This is strange. If you don't set any limit there shouldn't be any
> > > difference respect to the other case (without io-throttle patches).
> > > 
> > > At worst a small overhead given by the task_to_iothrottle(), under
> > > rcu_read_lock(). I'll repeat this test ASAP and see if I'll be able to
> > > reproduce this strange behaviour.
> > 
> > Ya, I also found this strange. At least in root group there should not be
> > any behavior change (at max one might expect little drop in throughput
> > because of extra code).
> 
> Hi Vivek,
> 
> I'm not able to reproduce the strange behaviour above.
> 
> Which commands are you running exactly? is the system isolated (stupid
> question) no cron or background tasks doing IO during the tests?
> 
> Following the script I've used:
> 
> $ cat test.sh
> #!/bin/sh
> echo 3 > /proc/sys/vm/drop_caches
> ionice -c 1 -n 0 dd if=bigfile1 of=/dev/null bs=1M 2>&1 | sed "s/$.*$/RT: \1/" &
> cat /proc/$!/cgroup | sed "s/$.*$/RT: \1/"
> ionice -c 2 -n 7 dd if=bigfile2 of=/dev/null bs=1M 2>&1 | sed "s/$.*$/BE: \1/" &
> cat /proc/$!/cgroup | sed "s/$.*$/BE: \1/"
> for i in 1 2; do
> 	wait
> done
> 
> And the results on my PC:
> 
> 2.6.30-rc4
> ~~~~~~~~~~
> $ sudo sh test.sh | sort
> BE: 234+0 records in
> BE: 234+0 records out
> BE: 245366784 bytes (245 MB) copied, 21.3406 s, 11.5 MB/s
> RT: 234+0 records in
> RT: 234+0 records out
> RT: 245366784 bytes (245 MB) copied, 11.989 s, 20.5 MB/s
> $ sudo sh test.sh | sort
> BE: 234+0 records in
> BE: 234+0 records out
> BE: 245366784 bytes (245 MB) copied, 23.4436 s, 10.5 MB/s
> RT: 234+0 records in
> RT: 234+0 records out
> RT: 245366784 bytes (245 MB) copied, 11.9555 s, 20.5 MB/s
> $ sudo sh test.sh | sort
> BE: 234+0 records in
> BE: 234+0 records out
> BE: 245366784 bytes (245 MB) copied, 21.622 s, 11.3 MB/s
> RT: 234+0 records in
> RT: 234+0 records out
> RT: 245366784 bytes (245 MB) copied, 11.9856 s, 20.5 MB/s
> $ sudo sh test.sh | sort
> BE: 234+0 records in
> BE: 234+0 records out
> BE: 245366784 bytes (245 MB) copied, 21.5664 s, 11.4 MB/s
> RT: 234+0 records in
> RT: 234+0 records out
> RT: 245366784 bytes (245 MB) copied, 11.8522 s, 20.7 MB/s
> 
> 2.6.30-rc4 + io-throttle, no BW limit, both tasks in the root cgroup
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> $ sudo sh ./test.sh | sort
> BE: 234+0 records in
> BE: 234+0 records out
> BE: 245366784 bytes (245 MB) copied, 23.6739 s, 10.4 MB/s
> BE: cgroup 4:blockio:/
> RT: 234+0 records in
> RT: 234+0 records out
> RT: 245366784 bytes (245 MB) copied, 12.2853 s, 20.0 MB/s
> RT: 4:blockio:/
> $ sudo sh ./test.sh | sort
> BE: 234+0 records in
> BE: 234+0 records out
> BE: 245366784 bytes (245 MB) copied, 23.7483 s, 10.3 MB/s
> BE: cgroup 4:blockio:/
> RT: 234+0 records in
> RT: 234+0 records out
> RT: 245366784 bytes (245 MB) copied, 12.3597 s, 19.9 MB/s
> RT: 4:blockio:/
> $ sudo sh ./test.sh | sort
> BE: 234+0 records in
> BE: 234+0 records out
> BE: 245366784 bytes (245 MB) copied, 23.6843 s, 10.4 MB/s
> BE: cgroup 4:blockio:/
> RT: 234+0 records in
> RT: 234+0 records out
> RT: 245366784 bytes (245 MB) copied, 12.4886 s, 19.6 MB/s
> RT: 4:blockio:/
> $ sudo sh ./test.sh | sort
> BE: 234+0 records in
> BE: 234+0 records out
> BE: 245366784 bytes (245 MB) copied, 23.8621 s, 10.3 MB/s
> BE: cgroup 4:blockio:/
> RT: 234+0 records in
> RT: 234+0 records out
> RT: 245366784 bytes (245 MB) copied, 12.6737 s, 19.4 MB/s
> RT: 4:blockio:/
> 
> The difference seems to be just the expected overhead.

BTW, it is possible to reduce the io-throttle overhead even more for non
io-throttle users (also when CONFIG_CGROUP_IO_THROTTLE is enabled) using
the trick below.

2.6.30-rc4 + io-throttle + following patch, no BW limit, tasks in root cgroup
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
$ sudo sh test.sh | sort
BE: 234+0 records in
BE: 234+0 records out
BE: 245366784 bytes (245 MB) copied, 17.462 s, 14.1 MB/s
BE: 4:blockio:/
RT: 234+0 records in
RT: 234+0 records out
RT: 245366784 bytes (245 MB) copied, 11.7865 s, 20.8 MB/s
RT: 4:blockio:/
$ sudo sh test.sh | sort
BE: 234+0 records in
BE: 234+0 records out
BE: 245366784 bytes (245 MB) copied, 18.8375 s, 13.0 MB/s
BE: 4:blockio:/
RT: 234+0 records in
RT: 234+0 records out
RT: 245366784 bytes (245 MB) copied, 11.9148 s, 20.6 MB/s
RT: 4:blockio:/
$ sudo sh test.sh | sort
BE: 234+0 records in
BE: 234+0 records out
BE: 245366784 bytes (245 MB) copied, 19.6826 s, 12.5 MB/s
BE: 4:blockio:/
RT: 234+0 records in
RT: 234+0 records out
RT: 245366784 bytes (245 MB) copied, 11.8715 s, 20.7 MB/s
RT: 4:blockio:/
$ sudo sh test.sh | sort
BE: 234+0 records in
BE: 234+0 records out
BE: 245366784 bytes (245 MB) copied, 18.9152 s, 13.0 MB/s
BE: 4:blockio:/
RT: 234+0 records in
RT: 234+0 records out
RT: 245366784 bytes (245 MB) copied, 11.8925 s, 20.6 MB/s
RT: 4:blockio:/

[ To be applied on top of io-throttle v16 ]

Signed-off-by: Andrea Righi <righi.andrea@gmail.com>
---
 block/blk-io-throttle.c |   16 ++++++++++++++--
 1 files changed, 14 insertions(+), 2 deletions(-)

diff --git a/block/blk-io-throttle.c b/block/blk-io-throttle.c
index e2dfd24..8b45c71 100644
--- a/block/blk-io-throttle.c
+++ b/block/blk-io-throttle.c
@@ -131,6 +131,14 @@ struct iothrottle_node {
 	struct iothrottle_stat stat;
 };
 
+/*
+ * This is a trick to reduce the unneded overhead when io-throttle is not used
+ * at all. We use a counter of the io-throttle rules; if the counter is zero,
+ * we immediately return from the io-throttle hooks, without accounting IO and
+ * without checking if we need to apply some limiting rules.
+ */
+static atomic_t iothrottle_node_count __read_mostly;
+
 /**
  * struct iothrottle - throttling rules for a cgroup
  * @css: pointer to the cgroup state
@@ -193,6 +201,7 @@ static void iothrottle_insert_node(struct iothrottle *iot,
 {
 	WARN_ON_ONCE(!cgroup_is_locked());
 	list_add_rcu(&n->node, &iot->list);
+	atomic_inc(&iothrottle_node_count);
 }
 
 /*
@@ -214,6 +223,7 @@ iothrottle_delete_node(struct iothrottle *iot, struct iothrottle_node *n)
 {
 	WARN_ON_ONCE(!cgroup_is_locked());
 	list_del_rcu(&n->node);
+	atomic_dec(&iothrottle_node_count);
 }
 
 /*
@@ -250,8 +260,10 @@ static void iothrottle_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
 	 * reference to the list.
 	 */
 	if (!list_empty(&iot->list))
-		list_for_each_entry_safe(n, p, &iot->list, node)
+		list_for_each_entry_safe(n, p, &iot->list, node) {
 			kfree(n);
+			atomic_dec(&iothrottle_node_count);
+		}
 	kfree(iot);
 }
 
@@ -836,7 +848,7 @@ cgroup_io_throttle(struct bio *bio, struct block_device *bdev, ssize_t bytes)
 	unsigned long long sleep;
 	int type, can_sleep = 1;
 
-	if (iothrottle_disabled())
+	if (iothrottle_disabled() || !atomic_read(&iothrottle_node_count))
 		return 0;
 	if (unlikely(!bdev))
 		return 0;
