🚀 go-pugleaf

In article <Bso916.7uE@ie.utoronto.ca>, andy@ie.utoronto.ca (Andy Sun) writes:
>
>Not long ago someone posted something on how newfs and mkfs ignore
>the inode option (-i) and apprently this problem has existed for a long
>time. Then someone from DEC respond as follows: 
>
>>From: schedler@src.dec.com (Richard Schedler)
>>Subject: Re: Will DEC ever fix newfs/mkfs ??
>>
>>Maybe this will help with your news partitions; it's what I use:
>>
>>    newfs -v -n -i 512 -b 4096 -f 512 -c 8 /dev/rrz1c rz58
>>
>>This will give you 512 bytes per inode, 4k blocks and 512 byte fragments.
>>The trick it to decrease the number of cylinders per group to 8.  
>
>Does anybody know if there's any side-effect in doing the above?

Taking the options one at a time:

-v   Instructs newfs to print out its actions, including the
     parameters passed to mkfs.  This will make it ever so
     slightly more verbose than normal.
-n   Prevents the bootstrap program from being installed.  Well,
     since you're using it for News, it seems unlikely that you'll
     the boot strap program.
-i   Number of bytes per inode.  When it works, this tells newfs
     what the expected average file size is, so it can try to
     allocate as many inodes as necessary.
-b   File system block size.  If you really expect the majority
     of news articles to fit into one 4096 byte block, then this
     is a reasonable block size.  It might be worth examining your
     existing News tree to see what the median and average sizes
     really are.
-f   The fragment size.  What do the small files look like?  Can
     you squeeze a little more space out of it by using a small
     fragment size?  Will you end up doing more I/O because of the
     smaller fragment size.
-c   The number of cylinders per group.  The smaller the cylinder
     group, the more cylinder groups there are.  As you have more
     cylinder groups you have more inodes.  It's part of the trade
     off of giving up capacity for more potential files.

>Or
>can someone provide me with pointers to any reference material?

The newfs(8) manual page will help some.  The fs(5) manual page
may also help some.  In the ULTRIX Supplementary Documentation
is the original paper descrbing the Fast File System; "A Fast
File System for UNIX".

>We have an RZ58 with /dev/rz1g as the news partition.

The basic parameters are about right.  I would look at how many
inodes you're using relative to the amount of space you're using
in your current News file system (df -i).  You want the two to
be reasonably close with enough ceiling in both to grow before
you start expiring news more often.

>Before I blindly follow
>the above, I would like to know the mechanism to arrive at the above
>numbers.

The command given assumes that New articles are small and that lots
of files are needed relative to the amount of space being used.  This
is probably true.  You do want to look closely at the median file
size to get a good estimate of how many inodes you need.  I haven't
looked closely at a News tree lately, but I believe the median file
size is larger than 512 bytes.  Some people signatures use up a large
part of that 512 and the headers may use the rest.  That doesn't leave
much for an article.

Also, over quoting of previous articles is probably pushing up the
median size.

>Andy
>-- 
>Andy Sun (andy@ie.utoronto.ca)		4 Taddle Creek Road, Toronto
>
--
Alan Rollow				alan@nabeth.cxo.dec.com