home | list info | list archive | date index | thread index

Re: [OCLUG-Tech] optimal choice for filesystem type for SSD drive doing big builds?

  • Subject: Re: [OCLUG-Tech] optimal choice for filesystem type for SSD drive doing big builds?
  • From: Stephen Gregory <oclug [ at ] kernelpanic [ dot ] ca>
  • Date: Mon, 11 Jan 2016 23:55:15 -0500
> On 16/01/11, Robert P. J. Day wrote:
>>
>>   normally, if i was interested in longevity, i'd put the *source* for
>> all my builds on the SSD, and leave the generation of build artifacts
>> on the regular drive. but at this point, i don't really care.

Given the guesstimates in this article you likely won't wear out your SSD:

http://www.anandtech.com/show/7173/samsung-ssd-840-evo-review-120gb-250gb-500gb-750gb-1tb-models-tested/3

That is specific to the Samsung, but they are all using similar tech
so the numbers are should be similar. TLDR: @100GB/day a 500 MB drive
should last 15 years.

>>   under these circumstances, is there an optimal choice for filesystem
>> type given what i'll be doing on that drive? i'm running fedora 23 so
>> ext4 is the simple and obvious choice, but does anyone have experience
>> with a different filesystem that would be preferable?

The CentOS 7 default is xfs. The Ubuntu 14.04 default is ext4. The
only difference I have found is that xfs can re-size a filesystem over
16TiB. (Ext4 on Ubuntu 14.04 can resize up to 16TiB, and create, but
not resize beyond 16TiB.) Performance wise they are each faster then
the other in different edge cases. As the Fedora default is ext4, use
ext4. Also ext4 is possibly the best debugged FS in the world.

As RGB pointed out btrfs is still in the not-quite-trusted stage.
Although I know people who claim they use it in production (and they
are crazy). IMHO the advantages to btrfs are better suited to large
multi-user filesystems anyway. Stuff like full data checksums to
detect silent bit rot, snapshots, efficient data replication and
backup, deduplication, and fully integrated raid5/6 (pending).

-- 
sg