Veritas File System 5.0 Release Notes, HP-UX 11i v3, First Edition, May 2008

Workaround
Avoid large offsets while using mmap.
• Problem
System hang can occur on systems that have less than 2GB cache memory.
Workaround
VxFS allocates a fixed amount of memory. You can change the amount of memory allocated
using the tunables vx_ninode and vxfs_bc_bufhwm. After long hours of operation,
low-memory systems (that is, systems having less physical RAM), may slow down or hang
due to memory pressure. To alleviate this problem, lower the values of vx_ninode and
vxfs_bc_bufhwm to limit VxFS memory consumption.
VxFS consumes nearly 10% of total physical memory for the default value of vx_ninode.
If the system has 512 MB of memory, VxFS will require up to 52 MB to store the VxFS inode
cache. Therefore, it is necessary to tune the tunables according to the type of load on the
system.
• Problem
100% full file system cannot be resized.
Workaround
In some circumstances, the fsadm command cannot resize a 100% full file system due to
lack of space for updating structural information. Check VxFS file systems on a regular basis
and increase their size if they approach 100% capacity. If a file system is busy or too
fragmented, the resize operation fails.
• Problem
Setting max_thread_proc to a low value may result in a system hang.
Workaround
If the value of the tunable max_thread_proc is less than 1100, the value will be
automatically set to 1100 when VxFS is installed. This value must be maintained at or above
1100.
• Problem
DMAPI dm_get_dirattrs may skip directory entries if user buffer is not large enough.
The vx_hsm_get_dirattr() reads directory entries 8 KB at a time. If the user buffer is
not large enough to hold the combined entries corresponding file statistics information (up
to ~114kb), then the next call to vx_hsm_get_dirattr() will continue from the wrong
offset within the directory, resulting in some directory entries being skipped.
Workaround
No error is displayed. The expectation is to continue from the last directory entry information
that could fit into the user’s buffer. However, the next directory read will begin 8 KB farther
down the directory. As a workaround, this issue can be avoided by passing a sufficiently
large user buffer. User buffer must be big enough to hold an entire directory block worth
of dm_stat_t entries, as one full directory block must be processed per invocation. For a
32 bit kernel, it works out to be about 116 KB for an 8 KB block size file system, so 128 KB
is used. For a 64 bit kernel, more than 128 KB is required because while converting a call
from a 32 bit application, more than the 128 KB is allocated, passed in by the application, to
take care of increase in structure size.
• Problem
Disk drives configured to use a write-back cache, or disk arrays configured with a volatile
write-back cache, exhibit data integrity problems.
Workaround
18 Veritas File System 5.0 Release Notes