[lustre-discuss] What does your (next) MDT look like?

Ben Evans bevans at cray.com
Fri Feb 23 06:58:51 PST 2018

Slightly left-field question for MDTs, will we enable data on inode for
really tiny files in ldiskfs?


On 2/22/18, 2:02 PM, "lustre-discuss on behalf of Dilger, Andreas"
<lustre-discuss-bounces at lists.lustre.org on behalf of
andreas.dilger at intel.com> wrote:

>On Feb 6, 2018, at 10:32, E.S. Rosenberg <esr+lustre at mail.hebrew.edu>
>> Hello fellow Lustre users :)
>> Since I didn't want to take the "size of MDT, inode count, inode size"
>>thread too far off-topic I'm starting a new thread.
>> I'm curious how many people are using SSD MDTs?
>> Also how practical is such a thing in a 2.11.x Data On MDT scenario?
>> Is using some type of mix between HDD and SSD storage for MDTs
>> Does SSD vs HDD have an effect as far as ldiskfs vs zfs?
>It is worthwhile to mention that using DoM is going to be a lot easier
>with ZFS in a "fluid" usage environment than it will be with ldiskfs.
>The ZFS MDTs do not have pre-allocated inode/data separation, so enabling
>DoM will just mean you can put fewer inodes on the MDT if you put more
>data there.  With ldiskfs you have to decide this ratio at format time.
>The drawback is that ZFS is somewhat slower for metadata than ldiskfs,
>though it has improved in 2.10 significantly.
>Cheers, Andreas
>Andreas Dilger
>Lustre Principal Architect
>Intel Corporation
>lustre-discuss mailing list
>lustre-discuss at lists.lustre.org

More information about the lustre-discuss mailing list