xmemfs(7FS) File Systems xmemfs(7FS)NAMExmemfs - extended memory file system
SYNOPSIS
#include <sys/mount.h>
mount(special_file, directory, MS_DATA, "xmemfs", dataptr, datalen);
INTERFACE LEVEL
This interface is obsolete. If you require support for large physical
memory, please migrate to 64-bit platform support.
DESCRIPTION
The xmemfs file system is an extended memory file system that provides
an efficient mechanism for managing and accessing physical memory that
exceeds 4 Gbytes in size. Currently, the xmemfs file system is sup‐
ported on IA32 architecture systems only.
The Physical Address Extension (PAE) is the xmemfs internal processor
feature that enables a 36-bit physical memory address that supports up
to 64 Gbytes of physical memory. Once mounted, the xmemfs file system
provides standard file operations and semantics on directories and
regular files only. Because xmemfs does not allow execute permissions
to be set on regular files, execution of object files is prevented.
With xmemfs, the special_file argument, (typically the device on which
file systems reside), is ignored and serves only as a placeholder. File
data and metadata in xmemfs are always memory-resident. The dataptr
argument must (at a minimum) contain the required size specific option.
See mount_xmemfs(1M) for more information.
Because xmemfs is a memory-based file system, files and directories
that are created are not persistent across reboots or unmounts.
EXTENDED DESCRIPTION
To mount the xmemfs file system, do the following: mount -Fxmemfs
-osize=4g xmem directory
You can also mount a xmemfs file system on /xmem at multi-user startup
time prior to physical memory becoming fragmented. To do this, add the
following line to your /etc/vfstab file: xmem
- /xmem xmemfs - yes largebsize,size=4g
The xmemfs file system is expressly designed for performance-driven
applications (for example, RDBMS) that require large amounts of physi‐
cal memory. The xmemfs file system provides file system semantics to
manage and access extended memory spaces that exceed 4 Gbytes. From an
application perspective, extended memory under the control of a mounted
xmemfs file system is viewed as a single, large memory pool that can be
partitioned as needed through file creation. You can obtain windows
into each memory partition by using mmap(2).
Memory controlled by xmemfs can be partitioned by creating files of
the required size in the file system. The xmemfs file system allocates
sufficient block-sized memory pages for a file based on the file's
size. Files can be created using any standard file utility, including
mkfile(1M) and dd(1M). The xmemfs file system optimizes the creation
of large files that initially contain all zeroes by allocating memory
pages for the file 'hole' that is created by writing beyond the end of
file.
If sufficient xmemfs extended memory is available, an application can
quickly create an 8 Gbyte file in the xmemfs file system by using
llseek(2) to offset 8GB-1 and then write(2) a one-byte buffer contain‐
ing zero. With xmemfs, you can share and protect partitioned memory by
setting appropriate file permissions. To avoid wasting memory
resources, (especially with the -largebsize option specified), newly
created option-specified files should be a multiple of the block size
of the xmemfs file system. Creation of many small files is strongly
discouraged. See statvfs(2) for information on determining file system
block sizes.
The xmemfs file system should only be used with performance-driven
applications that require quick access to large amounts of physical
memory. Using xmemfs for other applications may result in non-optimal
use of system resources and possible system performance degradation.
To maximize xmemfs ability to access a file's extended memory parti‐
tion, use mmap(2). The initial mmap(2) call enables the system to
assign a map size containing as much memory as an application may
actively access at any time. The map size is constrained by the appli‐
cation's virtual address space, (usually a maximum of 3 Gbytes on
machines with more than 4 Gbytes of physical memory). To access
extended memory that is not contained in the existing mapping, use
mmap(2) with the -MAP_FIXED flag to remap a window within the address
range returned by the initial mmap call.
ATTRIBUTES
See attributes(5) for descriptions of the following attributes:
┌─────────────────────────────┬─────────────────────────────┐
│ATTRIBUTE TYPE │ATTRIBUTE VALUE │
├─────────────────────────────┼─────────────────────────────┤
│Architecture │i386 │
├─────────────────────────────┼─────────────────────────────┤
│Interface Stability │Obsolete │
└─────────────────────────────┴─────────────────────────────┘
SEE ALSOdf(1M), mount(1M), mount_xmemfs(1M), mmap(2), mount(2), statvfs(2),
umount(2), vfstab(4)NOTES
Support for xmemfs may be removed from a future Solaris release.
DIAGNOSTICS
If the xmemfs file system runs out of space, the following message is
displayed in the console indicating that there is insufficient memory
to satisfy a write(2) request:
directory: File system full, no memory
WARNINGS
Files and directories on an xmemfs file system are not preserved across
reboots or unmounts.
SunOS 5.10 9 March 2006 xmemfs(7FS)
