Working with hugetlbfs huge-page support

6.18 LPAR mode z/VM guest KVM guest

Typical tasks for working with hugetlbfs huge-page support include reading the current number of huge pages, changing the number of huge pages, and display information about available huge pages.

About this task

The huge page memory can be used through mmap() or SysV shared memory system calls. More detailed information can be found in the Linux® kernel source tree under Documentation/vm/hugetlbpage.txt, including implementation examples.

Your database product might support huge-page memory. See your database documentation to find out if and how it can be configured to use huge-page memory.

Depending on your version of Java™, you might require specific options to make a Java program use the huge-page feature. For IBM® SDK, Java Technology Edition 7, specify the -Xlp option. If you use the SysV shared memory interface, which includes java -Xlp, you must adjust the shared memory allocation limits to match the workload requirements. Use the following sysctl attributes:

/proc/sys/kernel/shmall: Defines the global maximum amount of shared memory for all processes, specified in number of 4 KB pages.
/proc/sys/kernel/shmmax: Defines the maximum amount of shared memory per process, specified in number of Bytes.

For example, the following commands would set both limits to 20 GB:

# echo 5242880 > /proc/sys/kernel/shmall
# echo 21474836480 > /proc/sys/kernel/shmmax

Procedure

Specify the hugepages= kernel parameter with the number of huge pages to be allocated at boot time. To read the current number of default-sized huge pages, issue:
```
# cat /proc/sys/vm/nr_hugepages
```
To change the number of default-sized huge pages dynamically during runtime, write to procfs:
```
# echo 20 > /proc/sys/vm/nr_hugepages
```
If there is not enough contiguous memory available to fulfill the request, the maximum possible number of huge pages are reserved.

To obtain information about the number of default-sized huge pages currently available and the default huge-page size, issue:

# cat /proc/meminfo

...
HugePages_Total: 20
HugePages_Free: 14
HugePages_Rsvd: 0
HugePages_Surp: 0
...
Hugepagesize: 1024 kB
...

To adjust characteristics of a huge-page pool, when more than one pool exists, use the sysfs attributes of the pool.
These can be found under
```
/sys/kernel/mm/hugepages/hugepages-<size>/nr_hugepages
```
Where <size> is the page size in KB.

Example

To allocate 2 GB huge pages:

Specify 2 GB huge pages and pre-allocate them to the page pool at boot time. Use the following kernel boot parameters:
```
default_hugepagesz=2G hugepagesz=2G hugepages=4
```

After booting, read /proc/meminfo to see information about the amount of huge pages currently available and the huge-page size:

cat /proc/meminfo
...
HugePages_Total: 4
HugePages_Free: 4
HugePages_Rsvd: 0
HugePages_Surp: 0
Hugepagesize: 2097152 kB
...