I'd suggest to start with regular (non-SPDK) configuration and
deploy test cluster. Then do some benchmarking against it and
check if nvme drive is the actual bottleneck. I doubt it is
though. I did some experiments a while ago and didn't see any
benefit from SPDK in my case - probably due to bottlenecks were
somewhere else.
Hey, sorry if the question doesnt really make a lot of sense I
am talking from almost complete ignorace of the topic, but there
is not a lof info about it
I am planning about creating a cluster with 7~10 NL-SAS HDD and
1 nmve per host.
The nmve would be used as rocksDB and journal for each osd (hdd
block device) but I am worried about that being a bottleneck as
all the osds would share the same device.
I've been reading about spdk for nmve devices and I have seen
that blueStore configuration supports it
(http://docs.ceph.com/docs/master/rados/configuration/bluestore-config-ref/#spdk-usage)
but I would like the know the current status of the spdk and if
I could use it for rocksDB device and not for block device
I've been reading also about RDMA and I would like to know if I
could use it in this scenario, all I have found was using hole
nmve devices for block and RocksDB
I would really apreciate if someone could introduce me about
this topic, its really interesting but also confusing at the
same time.
Thanks a lot!
--
Jorge Pinilla López
jorpilo@xxxxxxxxx
Estudiante de ingenieria informática
Becario del area de sistemas (SICUZ)
Universidad de Zaragoza
PGP-KeyID:
A34331932EBC715A
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
