MemFS limitations ?

[mobidyc]

Hello,

I want to mount a 20G memfs, I have enough free memory (63G) to do this but I always have an error message.

# kmeminfo |grep Free
Free memory = 16610061 63.4g 40%

# mount -F memfs -o size=20g /mnt
mount: on /mnt : No space left on device

It works fine for a 15G FS:

# mount -F memfs -o size=15g /mnt
# bdf /mnt
Filesystem kbytes used avail %used Mounted on
memfs 15390688 4 13851612 0% /mnt

I've checked maxdsiz_64bit and UFS_MAXDEVBLK (in sys.fs.h), I should not be limited.

Please, any advice ?

Regards,
Cedrick Gaillard

[Dennis Handly]

This is a duplicate of your previous thread. Perhaps it is time to contact the Response Center with your question.
http://forums.itrc.hp.com/service/forums/questionanswer.do?threadId=1354172

[Ganesan R]

Hi,

Have you read this document?

http://docs.hp.com/en/5992-4721/5992-4721.pdf

It says,

memfs_swapmax_pct
The memfs_swapmax_pct tunable defines the maximum amount of system swap
space that can be used by all MemFS instances in a system as percentage of total system
swap configured in a system. The tunable does not reserve or assure the swap space
of specified size; it just assures that the swap space usage by MemFS does not grow
beyond the specified size. The value of this tunable can be as low as 0% of system swap
or as high as 80% of system swap. The default value is 50% of system swap.

check the value of memfs_swapmax_pct

[mobidyc]

Hello Dennis,

already done, no answers for a week.

Regards,
Cedrick Gaillard

[mobidyc]

Hello Ganesan R,

I have read this document and memfs_swapmax_pct tunable must not limit the MemFS size, It only limit swapping when there is memory pressure.

Anyway I'm on 11iv2 and this document is applicable to 11iv2 only.

the corresponding tunable parameter for 11iv2 is memfs_bufcache_swappct.

even if this parameter was limitating, It is bigger enough to support actual swapping + 90G:
# swapinfo -tam
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 32000 0 32000 0% 0 - 1 /dev/vg00/lvol2
dev 64000 0 64000 0% 0 - 0 /dev/vg01/lvolswap
reserve - 75767 -75767
memory 163579 95857 67722 59%
total 259579 171624 87955 66% - 0 -

# kctune memfs_bufcache_swappct
Tunable Value Expression Changes
memfs_bufcache_swappct 50 50 Immed


Regards,
Cedrick Gaillard

[Ganesan R]

Hi,

man page of mount_memfs says,

The size of MemFS file systems is limited by UFS_MAXDEVBLK, defined in ,

http://docs.hp.com/en/B3921-60631/mount_memfs.1M.html

[mobidyc]

Ganesan R,

please, read my first post:
I've checked maxdsiz_64bit and UFS_MAXDEVBLK (in sys.fs.h), I should not be limited.

UFS_MAXDEVBLK is at its default value.

# grep UFS_MAXDEVBLK /usr/include/sys/fs.h
* (fragments) which are supported for the file system. UFS_MAXDEVBLK is
* the maximum device block number for a UFS file system. UFS_MAXDEVBLK
#define UFS_MAXDEVBLK 0x0FFFFFFE /* Same as MAX_DBD_DATA */
#define UFS_MAXFSBLK(fs) dbtofsb((fs), UFS_MAXDEVBLK)

Regards,
Cedrick Gaillard

[ManojK_1]

Hi,

Can you please try with the below command
/sbin/fs/memfs/mount -F memfs -o size=20g /mnt

and Please post the output and verify in syslog any error thrown there.

Thanks and Regards,
Manoj K

[Aneesh Mohan]

Hi,

fyi.

MemFS file system size to be created is large (greater than 2 GB), the tunable maxdsiz_64bit must be tuned appropriately.


memfs_bufcache_metapct
is the maximum percentage of buffer cache which can accommodate
MemFS metadata of all MemFS instances. The default value is 10. The
memfs_bufcache_metapct tunable cannot be less than 10 and cannot be
greater than 50. Tuning memfs_bufcache_metapct to a value as high as 50
may be necessary if the system primarily uses large MemFS filesystems. The
larger memfs_bufcache_metapct is, the more of the buffer cache will be
reserved for MemFS metadata pages and hence will allow creation of more
and larger MemFS filesystem instances, but impact the performance of
other filesystems that use the buffer cache.


Regards,
Aneesh

[mobidyc]

Hello,

the problem was the dbc_max_pct not large enough.

thanks all for your answers.

Cheers,
Cedrick Gaillard