OpenVZ Forum: Devel » [RFC][PATCH] Containers: Pagecache accounting and control subsystem (v1)

Home » Mailing lists » Devel » [RFC][PATCH] Containers: Pagecache accounting and control subsystem (v1)

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

[RFC][PATCH] Containers: Pagecache accounting and control subsystem (v1) [message #10524]

Tue, 20 February 2007 13:52

Vaidyanathan Srinivas is currently offline

Vaidyanathan Srinivas
Messages: 49
Registered: February 2007

Member

This patch adds pagecache accounting and control on top of
Paul's container subsystem v7 posted at
http://lkml.org/lkml/2007/2/12/88

Comments, suggestions and criticisms are welcome.

Features:
--------
* New subsystem called 'pagecache_acct' is registered with containers
* Container pointer is added to struct address_space to keep track of
associated container
* In filemap.c and swap_state.c, the corresponding container's
pagecache_acct subsystem is charged and uncharged whenever a new
page is added or removed from pagecache
* The accounting number include pages in swap cache and filesystem
buffer pages apart from pagecache, basically everything under
NR_FILE_PAGES is counted as pagecache. However this excluded
mapped and anonymous pages
* Limits on pagecache can be set by echo 100000 > pagecache_limit on
the /container file system. The unit is in kilobytes
* If the pagecache utilisation limit is exceeded, pagecache reclaim
code is invoked to recover dirty and clean pagecache pages only.

Advantages:
-----------
* Does not add container pointers in struct page

Limitations:
-----------
* Code is not safe for container deletion/task migration
* Pagecache page reclaim needs performance improvements
* Global LRU is churned in search of pagecache pages

Usage:
-----

* Add patch on top of Paul container (v7) at kernel version 2.6.20
* Enable CONFIG_CONTAINER_PAGECACHE_ACCT in 'General Setup'
* Boot new kernel
* Mount container filesystem
mount -t container /container
cd /container
* Create new container
mkdir mybox
cd /container/mybox
* Add current shell to container
echo $$ > tasks
* There are two files pagecache_usage and pagecache_limit
* In order to set limit, echo value in kilobytes to pagecache_limit
echo 100000 > pagecache_limit
#This would set 100MB limit on pagecache usage
* Trash the system from current shell using scp/cp/dd/tar etc
* Watch pagecache_usage and /proc/meminfo to verify behavior

* Only unmapped pagecache data will be accounted and controlled.
These are memory used by cp, scp, tar etc. While file mmap will
be controlled by Balbir's RSS controller.

ToDo:
----

* Merge with container RSS controller and eliminate redundant code
* Support and test task migration and container deletion
* Review reclaim performance
* Optimise page reclaim

Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
---
include/linux/fs.h | 6
include/linux/pagecache_acct.h | 52 +++++
init/Kconfig | 7
kernel/Makefile | 1
kernel/pagecache_acct.c | 376 +++++++++++++++++++++++++++++++++++++++++
mm/filemap.c | 8
mm/vmscan.c | 76 ++++++++
7 files changed, 524 insertions(+), 2 deletions(-)

--- linux-2.6.20.orig/include/linux/fs.h
+++ linux-2.6.20/include/linux/fs.h
@@ -447,6 +447,12 @@ struct address_space {
spinlock_t private_lock; /* for use by the address_space */
struct list_head private_list; /* ditto */
struct address_space *assoc_mapping; /* ditto */
+
+#ifdef CONFIG_CONTAINER_PAGECACHE_ACCT
+ struct container *container; /* Charge page to the right container
+ using page->mapping */
+#endif
+
} __attribute__((aligned(sizeof(long))));
/*
* On most architectures that alignment is already the case; but
--- /dev/null
+++ linux-2.6.20/include/linux/pagecache_acct.h
@@ -0,0 +1,52 @@
+/*
+ * Pagecache controller - "Account and control pagecache usage"
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * � Copyright IBM Corporation, 2007
+ *
+ * Author: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
+ *
+ */
+
+#ifndef _LINUX_PAGECACHE_ACCT_H
+#define _LINUX_PAGECACHE_ACCT_H
+
+#include <linux/container.h>
+
+#ifdef CONFIG_CONTAINER_PAGECACHE_ACCT
+extern void pagecache_acct_init_page_ptr(struct page *page);
+extern void pagecache_acct_charge(struct page *page);
+extern void pagecache_acct_uncharge(struct page *page);
+extern int pagecache_acct_page_overlimit(struct page *page);
+extern int pagecache_acct_mapping_overlimit(struct address_space *mapping);
+extern int pagecache_acct_cont_overlimit(struct container *cont);
+extern int pagecache_acct_shrink_used(unsigned long nr_pages);
+#else
+static inline void pagecache_acct_init_page_ptr(struct page *page) {}
+static inline void pagecache_acct_charge(struct page *page) {}
+static inline void pagecache_acct_uncharge(struct page *page) {}
+static inline int pagecache_acct_page_overlimit(
+ struct page *page) { return 0; }
+static inline int pagecache_acct_mapping_overlimit(
+ struct address_space *mapping) { return 0; }
+static inline int pagecache_acct_cont_overlimit(
+ struct container *cont) { retrun 0; }
+static inline int pagecache_acct_shrink_used(
+ unsigned long nr_pages) { retrun 0; }
+#endif
+
+#endif
+
--- linux-2.6.20.orig/init/Kconfig
+++ linux-2.6.20/init/Kconfig
@@ -306,6 +306,13 @@ config CONTAINER_NS
for instance virtual servers and checkpoint/restart
jobs.

+config CONTAINER_PAGECACHE_ACCT
+ bool "Simple PageCache accounting container subsystem"
+ select CONTAINERS
+ help
+ Provides a simple Resource Controller for monitoring the
+ total pagecache consumed by the tasks in a container
+
config RELAY
bool "Kernel->user space relay support (formerly relayfs)"
help
--- linux-2.6.20.orig/kernel/Makefile
+++ linux-2.6.20/kernel/Makefile
@@ -40,6 +40,7 @@ obj-$(CONFIG_CONTAINERS) += container.o
obj-$(CONFIG_CPUSETS) += cpuset.o
obj-$(CONFIG_CONTAINER_CPUACCT) += cpu_acct.o
obj-$(CONFIG_CONTAINER_NS) += ns_container.o
+obj-$(CONFIG_CONTAINER_PAGECACHE_ACCT) += pagecache_acct.o
obj-$(CONFIG_IKCONFIG) += configs.o
obj-$(CONFIG_STOP_MACHINE) += stop_machine.o
obj-$(CONFIG_AUDIT) += audit.o auditfilter.o
--- /dev/null
+++ linux-2.6.20/kernel/pagecache_acct.c
@@ -0,0 +1,376 @@
+
+/*
+ * Pagecache controller - "Account and control pagecache usage"
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * � Copyright IBM Corporation, 2007
+ *
+ * Author: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/container.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/mm_types.h>
+#include <linux/uaccess.h>
+#include <asm/div64.h>
+#include <linux/klog.h>
+#include <linux/pagecache_acct.h>
+
+/*
+ * Convert unit from pages to kilobytes
+ */
+#define K(x) ((x) << (PAGE_SHIFT - 10))
+/*
+ * Convert unit from kilobytes to pages
+ */
+#define K_to_pages(x) ((x) >> (PAGE_SHIFT -10))
+
+/* Limits for user string */
+
+#define MAX_LIMIT_STRING 25
+
+/* nr_pages above limit to start reclaim */
+
+#define NR_PAGES_RECLAIM_THRESHOLD 64
+
+struct pagecache_acct {
+ struct container_subsys_state css;
+ spinlock_t lock;
+ atomic_t count; /*Pagecache pages added*/
+ atomic_t removed_count; /*Pagecache pages removed*/
+ unsigned int limit; /* Pagecache usage limit in kiB */
+};
+
+/*Failure counters for debugging*/
+static atomic_t failed_count; /*Page charge failures?*/
+static atomic_t failed_removed_count; /*Page uncharge failure?*/
+static atomic_t reclaim_count; /*Overlimit direct page reclaim run count */
+
+static struct container_subsys pagecache_acct_subsys;
+
+static inline struct pagecache_acct *container_pca(struct container *cont)
+{
+ return container_of(
+ container_subsys_state(cont, &pagecache_acct_subsys),
+ struct pagecache_acct, css);
+}
+
+static int pagecache_acct_create(struct container_subsys *ss,
+ struct container *cont)
+{
+ struct pagecache_acct *pca = kzalloc(sizeof(*pca), GFP_KERNEL);
+ if (!pca)
+ return -ENOMEM;
+ spin_lock_init(&pca->lock);
+ cont->subsys[pagecache_acct_subsys.subsys_id] = &pca->css;
+ return 0;
+}
+
+static void pagecache_acct_destroy(struct container_subsys *ss,
+ struct container *cont)
+{
+ kfree(container_pca(cont));
+}
+
+static unsigned int pagecache_get_usage(struct pagecache_acct *pca)
+{
+ unsigned int count, removed_count, pagecache_used_kB;
+
+ count = (unsigned int) atomic_read(&pca->count);
+ removed_count = (unsigned int) atomic_read(&pca->removed_count);
+ /* Take care of roll over in the counters */
+ if (count >= removed_count)
+ pagecache_used_kB = count - removed_count;
+ else
+ pagecache_used_kB = ~0UL - (removed_count - count) + 1;
+
+ /* Convert unit from pages into kB */
+ pagecache_used_kB = K(pagecache_used_kB);
+
+ return pagecache_used_kB;
+}
+
+sta ...

[ Show the rest of the message ]

Report message to a moderator