MemFS v2 û A Memory-based File System on HP-UX 11i v2

Figure 5: SDET Benchmark results
SDET Results
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
1
20
40
60
80
100
5
00
Concurrent workload
Transactions/sec
MemFS
VxFS
HFS
RAMdisk/VxFS
Linux/tmpfs
Linux/ext2fs
The following conclusions can be drawn:
• MemFS performs better on the SDET benchmark compared to disk based filesystems and
RAMdisk
• MemFS performs better on the SDET benchmark compared to Linux’s tmpfs on the same test
machine
MemFS mount/umount performance
Since MemFS needs to create and initialize all meta-data structures during a mount operation, there
will be a noticeable amount of time taken to create large MemFS instances. User process memory
whose size is equal to the size of the file system needs to be pre-allocated at mount time. In addition,
MemFS meta-data structures are stored in the buffer cache and the buffer cache resources are not
released until the MemFS instance is unmounted. Therefore, it is recommended that the size of the
MemFS instance is chosen appropriately as required during mount.
Similarly, the umount operation may take some time to complete for a large MemFS instance, as
MemFS needs to de-allocate all related structures and free up resources by traversing through the
entire buffer cache looking for buffers that need to in-validated.
Configuration Guidelines
A MemFS instance requires allocation of a memory area to be associated with it, whose size is equal
to the size of the MemFS instance. The mount command will fail if it cannot allocate enough memory.
Sufficient disk space must be configured to the system swap device for mount to succeed. If the
MemFS file system size to be created is large (greater than 2 GB), the tunable maxdsiz_64bit must be
tuned appropriately. This memory area is allocated in the user address space of a process that is
forked off the mount command for every mount.
The maximum number of MemFS file system instances supported in HP-UX 11iv2 is 64.