Hi!
On one of our deployments we have almost 30 million of files where the
master server has 16GB RAM from which 10GB are occupied by mfsmaster. We do
not observe any problems with metadata testing, although we could think of
an option of reducing the speed of tests. Now the system makes the whole
testing during one hour which by 10 million files means testing about 3,000
inodes per second.
For every save operation there is a separate save process initialized which
does the operation in the background, usually on different processor core so
it should not slow down the master itself. Of course saving metada for 30
million files takes some time (about 50 seconds) but we do not encounter any
problems during saving.
We think 6GB RAM may be too little for 10+ million files. What do you
exactly mean by 10+? Is it 10,5 million or maybe 15?
If you could also give us more detailed information, maybe some screens from
mfscgiserv?
Kind regards
Michał Borychowski
MooseFS Support Manager
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Gemius S.A.
ul. Wołoska 7, 02-672 Warszawa
Budynek MARS, klatka D
Tel.: +4822 874-41-00
Fax : +4822 874-41-01
From: Shen Guowen [mailto:sh...@ui...]
Sent: Friday, May 21, 2010 10:31 AM
To: moo...@li...
Cc: 'jiliang'; 'Gavin Huang'
Subject: [Moosefs-users] [MooseFS-user] Master server is overloaded for
storing metadate periodically
Hi moosefs developer,
The master server is storing metadata periodically. However, after testing
files increases up to 10+ million, this process is becoming so
time-consuming, that the master cannot provides services normally. it is
overloaded by metadate storing. And from the mfscgiserv, as the increasing
files writes into, the spent time of metadata storing is becoming longer.
Besides, during each fs_storeall(), much percentage of fsnodes and chunks
are unchanged, for that most operations are read. Maybe lots of repeated
saving items are not necessary, but I do not know how to reduce them.
I think it may will be better to refresh the metadata in another sub-master
server by combining the original metadata and changlogs, and remove the
periodical storing from the master.
Do you have any solutions or opinions about this issue? Thanks!
OS: CentOS 5.3
Hardware: CPU xeon 5410 RAM: 6GB
Version: mfs-1.6.15
--
Yours sincerely, Guowen Shen
|
