SUMMARY: How long needed for doing fsck

From: <cceckw_at_leonis.nus.sg> <cceckw_at_leonis.nus.sg>
Date: Tue, 2 Apr 1996 10:20:33 +0800 (SST)

I'd like to apologise for the late reply as I'd been out of office for the
past 3 weeks.

Altogether, I've received 4 replies but none of them contains any hard
figures. I'd like to thank the following people for their contributions:

Jari Tavi <jpt_at_namix.ENET.dec.com>
Alan Rollow <alan_at_nabeth.cxo.dec.com>
Chua Koon Teck <koonteck_at_singnet.com.sg>
Tim Mooney <mooney_at_dogbert.cc.ndsu.NoDak.edu>

------------------------------------------------------------------------
The original question was:
------------------------------------------------------------------------
Subject: How long needed for doing fsck?

Hi osf managers,

Please pardon me if this sounds trivial to some of you. We are thinking
of converting our mail file system from Advfs to UFS because Advfs is
putting too much load on cpu 0. However, we don't have any experience
with running UFS on such a dynamic file system and so we are rather
concerned with regards to how long would it take to do fsck. Moreover,
we are not able to get any system downtime to do our own testing. Any
rough estimates from anyone especially if you have lots of open files?
Our file system has got 24,000 files and we got to cater for the worst
case (ie all the files are open). Our system is DEC/7000 model 730
running DUNIX 3.2c.

Thank you all for your valuable help in advance.

-------------------------------------------------------------------------
Some of the replies were:
-------------------------------------------------------------------------

From: Tim Mooney <mooney_at_dogbert.cc.ndsu.NoDak.edu>
Subject: Re: How long needed for doing fsck?

When we last communicated, alpha-osf-managers-relay_at_sws1.ctd.ornl.gov wrote:
 
)Hi osf managers,
 
)Please pardon me if this sounds trivial to some of you. We are thinking
)of converting our mail file system from Advfs to UFS because Advfs is
)putting too much load on cpu 0.

According to reports I've seen in this list, the Advfs code with be symmetric
in Digital Unix 4.0, due to ship this month, so you might want to hold off on
that change. It could be that 4.0 alleviates your problem (of course, it could
create a bunch of new ones too... :-) ).

Tim

-- 
Tim Mooney                              mooney_at_toons.cc.ndsu.NoDak.edu
Information Technology Services         (701) 231-1076 (Voice)
Room 242-J6, IACC Bldg.                 (701) 231-8541 (FAX)
North Dakota State University, Fargo, ND 58105
---------------------------------------------------------------------------
From: Jari Tavi  11-Mar-1996 1343 <jpt_at_namix.ENET.dec.com>
Subject: Re: How long needed for doing fsck?  
	Hi,
	Giving any estimates would be very difficult without real life
	testing, mostly as time depends on several things:
		- what is i/o performance of system
		- what i/o controllers
		- how many disks
		- disk types
		- how many fsck's can be run in parallel
		- how many dile systems there are in total
	For example RZ29B performs almost twice the i/o operations
	compared to RZ28, and makes fsck significantly faster as long
	as there is free cpu time available.
	HSZ-disk controllers with latest firmware can give very good
	i/o rates specially if write back cache has been enabled etc.
	So, giving estimates would be dependent on too many details
	and is totally dependent on the environment you have. 
	If AdvFS funneling in V3.2 is problem, have you evaluated 
	possibility to move to V4.0 as AdvFS distributes its load
	over all cpu's in that version?
	regards,
			-jari
---------------------------------------------------------------------------	 
From: alan_at_nabeth.cxo.dec.com (Alan Rollow - Dr. File System's Home for Wayward Inodes.)
Subject: Re:  How long needed for doing fsck?
I believe there are four factors that affect the time needed
run fsck on a file system:
1.  Number of cylinder groups.
2.  Number of files in each group.
3.  Total size of the file system.
4.  Whether the file system needs a 2nd pass.
None of these really depend on how many files are open, just
how many files there are.  Since the writes to update directories,
inodes, etc are all synchronous the file system will tend to be
fairly consistent after a system failure.  Unless some data became
corrupted in the crash an open file is little different than a
closed done.  The difference will be that stale data in the
buffer cache won't have made it to disk and the contents of
a recently extended file may be "interesting". 
Some of the checks made are done cylinder group by cylinder group, 
and a larger file system has more cylinder groups.  Once in a group, 
checks have to be made each file; consistency of the inode and the 
directory structure.  I don't know if file system depth takes longer
to check than file system breath.
Virtual memory is used to keep track of the state of many of
the data structures.  As the file system gets larger, more
memory is needed.  I think it works out to around 1 MB of
memory to each GB of file system space.  This will tend to
only matter for very large file systems.
If the file system needs to be repaired after a system failure,
then a 2nd pass is needed and the time is approximately doubled.
--------------------------------------------------------------------------

Received on Tue Apr 02 1996 - 04:44:13 NZST