Re: [gentoo-user] Re: MBR partition

[Kerin Millar]

On 06/09/2014 13:54, Alan McKinnon wrote:
> On 06/09/2014 14:48, Dale wrote:
>> James wrote:
>>> Joseph <syscon780 <at> gmail.com> writes:
>>>
>>>> Thank you for the information.
>>>> I'll continue on Monday and let you know.  If it will not boot with sector
>>> starting at 2048, I will
>>>> re-partition /boot sda1 to start at 63.
>>>
>>> Take some time to research and reflect on your needs (desires?)
>>> about which file system to use. (ext 2,4) is always popular and safe.
>>> Some are very happy with BTRFS and there are many other interesting
>>> choices (ZFS, XFS, etc etc)......
>>>
>>> There is no best solution; but the EXT family offers tried and proven
>>> options. YMMV.
>>>
>>>
>>> hth,
>>> James
>>>
>>
>> I'm not sure if it is ZFS or XFS but I seem to recall one of those does
>> not like sudden shutdowns, such as a power failure.  Maybe that has
>> changed since I last tried whichever one it is that has that issue.  If
>> you have a UPS tho, shouldn't be so much of a problem, unless your power
>> supply goes out.
>
> XFS.
>
> It was designed by SGI for their video rendeing workstations back in the
> day and used very aggressive caching to get enormous throughput. It was
> also brilliant at dealing with directories containing thousands of small
> files - not unusual when dealing with video editing.
>
> However, it was also designed for environments where the power is
> guaranteed to never go off (which explains why they decided to go with
> such aggressive caching). If you use it in environments where powerouts
> are not guaranteed to not happen, well......

Well what? It's no less reliable than other filesystems that employ 
delayed allocation (any modern filesystem worth of note). Over recent 
years, I use both XFS and ext4 extensively in production and have found 
the former trumps the latter in reliability.

While I like them both, I predicate this assertion mainly on some of the 
silly bugs that I have seen crop up in the ext4 codebase and the 
unedifying commentary that has occasionally ensued. From reading the XFS 
list and my own experience, I have formed the opinion that the 
maintainers are more stringent in matters of QA and regression testing 
and more mature in matters of public debate. I also believe that 
regressions in stability are virtually unheard of, whereas regressions 
in performance are identified quickly and taken very seriously [1].

The worst thing I could say about XFS is that it was comparatively slow 
until the introduction of delayed logging (an idea taken from ext3). [2] 
[3]. Nowadays, it is on a par with ext4 and, in some cases, scales 
better. It is also one of the few filesystems besides ZFS that can 
dynamically allocate inodes.

>
>
>
> ZFS is the most resilient filesystem I've ever used, you can through the
> bucket and kitchen sink at it and it really doesn't give a shit (it just
> deals with it :-) )

While its design is intrinsically resilient - particularly its 
capability to protect against bitrot - I don't believe that ZFS on Linux 
is more reliable in practice than the filesystems included in the Linux 
kernel. Quite the contrary. Look at the issues labelled as "Bug" filed 
for both the SPL and ZFS projects. There are a considerable number of 
serious bugs that - to my mind - disqualify it for anything but hobbyist 
use and I take issue with the increasing tendency among the community to 
casually recommend it.

Here's my anecdotal experience of using it. My hosting company recently 
installed a dedicated backup server that was using ZFS on Linux. Its 
primary function was as an NFS server. It was very slow and repeatedly 
deadlocked under heavy load. On each occasion, the only remedy was for 
an engineer to perform a hard reboot. When I complained about it, I was 
told that they normally use FreeBSD but had opted for Linux because the 
former was not compatible with a fibre channel adapter that they needed 
to make use of. I then requested that the filesystem be changed to ext4, 
after which the server was rock solid.

Another experience I have is of helping someone resolve an issue where 
MySQL was not starting. It transpired that he was using ZFS and that it 
does not support native AIO. I supplied him with a workaround but 
sternly advised him to switch to a de-facto Linux filesystem if he 
valued his data and expected anything like decent performance from 
InnoDB. Speaking of which, XFS is a popular filesystem among 
knowledgeable MySQL hackers (such as Mark Callaghan) and DBAs alike.

For the time being, I think that there are other operating systems whose 
ZFS implementation is more robust.

--Kerin

[1] 
http://www.percona.com/blog/2012/03/15/ext4-vs-xfs-on-ssd/#comment-903938
[2] 
https://www.kernel.org/doc/Documentation/filesystems/xfs-delayed-logging-design.txt
[3] https://www.youtube.com/watch?v=FegjLbCnoBw