Bonjour,
In the past weeks a few mailing list threads[0][1][2] explored the problem of storing
billions of small objects in Ceph. There was great feedback (I learned at lot) and it
turns out the solution is a rather simple aggregation of the ideas that were suggested
during these discussions. It is described in detail here:
https://wiki.softwareheritage.org/wiki/A_practical_approach_to_efficiently_…
The next step will be to write and run benchmarks[3]. Although I'm convinced it is a
good solution, I've been wrong before and confirmation is required :-) Many thanks to
all the participants in the discussions on behalf of the Software Heritage project[4].
Cheers
[0] Storing 20 billions of immutable objects in Ceph, 75% <16KB
https://lists.ceph.io/hyperkitty/list/ceph-users@ceph.io/thread/JSG2TXKNXPX…
[1] Small RGW objects and RADOS 64KB minimun size
https://lists.ceph.io/hyperkitty/list/ceph-users@ceph.io/thread/AEMW6O7WVJF…
[2] Using RBD to pack billions of small files
https://lists.ceph.io/hyperkitty/list/ceph-users@ceph.io/thread/RHQ5ZCHJISX…
[3]
https://forge.softwareheritage.org/T3054
[4]
https://www.softwareheritage.org/
--
Loïc Dachary, Artisan Logiciel Libre