Re: [gentoo-user] Best file system for portage tree?

[Pandu Poluan <pandu@poluan.info>]

[-- Attachment #1: Type: text/plain, Size: 2295 bytes --]

On Mar 10, 2012 10:09 PM, "Pandu Poluan" <pandu@poluan.info> wrote:
>
>
> On Mar 10, 2012 8:33 PM, "Alex Schuster" <wonko@wonkology.org> wrote:
> >
> > Hi there!
> >
> > Is there an advantage in putting the portage tree on an extra partition?
> >
> > Currently, I'm using reiserfs, because I read that it is efficient when
> > using many small files. On the other hand I also heard that it tends to
> > get slower with every emerge --sync.
> >
> > Space is no longer an argument in these days, at least for my desktop
> > machine. But I would like to optimize for speed -- emerge -DputnVj
> > @world takes quite a while to calculate, I assume this is because so
many
> > ebuild files have to be accessed.
> >
> > Any tips on this? Does it make sense to use a special file system just
> > for the portage tree? What would be best? Would it help to re-create
this
> > file system from time to time in case it gets slower with every sync? Or
> > wouldn't I notice a difference if I just used a big ext4 partition for
> > all portage related stuff?
> >
> > Anyone using a compressed RAM file system for that? :)
> >
>
> This had been my burning question when I was deploying the company's
production server, and forced me to do some research:
>
> * reiserfs is amazingly fast for reads, but suffers on simultaneous writes
> * reiserfs does not have inode limits
> * reiserfs' notail affects performance greatly depending on the nature of
the system: I/O-bound (use notail) or CPU-bound (don't use notail)
> * reiserfs, if mounted without notail, is very space-efficient
>
> So, I end up with the following mix:
>
> * ext2 for /boot
> * reiserfs for /usr/portage and /var/tmp (RAM is at premium; can't use
tmpfs)
> * ext4 for everything else
>
> This cocktail has been serving me well. I don't need advanced filesystems
like ZFS, XFS, or btrfs, because my servers are virtualized, and the
advanced features (e.g., snapshot) is handled by the underlying hypervisor
(XenServer) and SAN Storage (we use NetApp).
>
> Rgds,

Okay, I did a mixup:

If the system is I/O-bound, *don't* use notail (saves on disk read/write).

If the system is CPU-bound, *use* notail (saves on having to 'unpack' the
tail from the metadata).

In my situation, the bottleneck is the SAN Storage, so I don't use notail.

Rgds,

[-- Attachment #2: Type: text/html, Size: 2916 bytes --]