|
|
RE: RE: qnx6f max file number
|
|
01/25/2011 11:20 AM
post82634
|
RE: RE: qnx6f max file number
-----Original Message-----
From: Mario Charest [mailto:community-noreply@qnx.com]
Sent: January-24-11 7:26 PM
To: general-filesystems
Subject: Re: RE: qnx6f max file number
>
>
> > -----Message d'origine-----
> > De : David Sarrazin [mailto:community-noreply@qnx.com]
> > Envoyé : 20 septembre 2010 09:55
> > À : general-filesystems
> > Objet : RE: qnx6f max file number
> >
> > Each inode requires 128 bytes, so 100M inodes is 12.8GB, plus the indirect
> > blocks. Finding a free inode in a mostly-full system would be time
> expensive
> > at that scale. What you are saying is that you expect your 2TB drive to be
> > filled up with 10KByte average-sized files. Having said that, I know that
> the
> > test group here has a 2TB server set up with 10M inodes.
> >
> That would be 20K files ;-) I exaggerated that number a bit to get a good
> idea. Files will be 400K so 5 millions would be closer to what we need.
>
> Can you define 'time expensive'? The hardware driver will operate almost 90%
> full, with around 100 files new files written and the 100 oldest files deleted
> .
Given mkqnx6fs -i12000000 on a 2Terabyte disk filled at 8% is it possible that a write of 20 files (20Meg) will cause
the filesystem to freeze for 10-20Sec ?
I would expect no. If you've used up most of the inodes, and have just a few free holes spread about in the inodes file
, then finding a free inode will incur a search through the file (a hint to where to start searching is stored). If
this was a new filesystem, the inodes file will be almost entirely free, and finding new entries will be very fast.
Does a KEV show what the "fsys_resmgr" threads in devb are doing?
David
_______________________________________________
General
http://community.qnx.com/sf/go/post82603
|
|
|
|
|