[pve-devel] ZFS over LIO iscsi
Gerhard W. Recher
gerhard.recher at net4sec.com
Tue Oct 24 16:38:46 CEST 2017
Mark,
I have a all NVMe 4 nodes cluster
rdma is still not working as expected, but tcp performance is great even
on bigger Block size(s)
rados bench -p rbd 60 write --no-cleanup -t 56 -b 256K -o 1M
Total time run: 60.008792
Total writes made: 633524
Write size: 262144
Object size: 1048576
Bandwidth (MB/sec): 2639.3
Stddev Bandwidth: 139.635
Max bandwidth (MB/sec): 2850.75
Min bandwidth (MB/sec): 2300.5
Average IOPS: 10557
Stddev IOPS: 558
Max IOPS: 11403
Min IOPS: 9202
Average Latency(s): 0.00530374
Stddev Latency(s): 0.00560477
Max latency(s): 0.113396
Min latency(s): 0.000747469
and 1M blocksize:
Total time run: 60.025703
Total writes made: 198400
Write size: 1048576
Object size: 1048576
Bandwidth (MB/sec): 3305.25
Stddev Bandwidth: 35.2218
Max bandwidth (MB/sec): 3379
Min bandwidth (MB/sec): 3229
Average IOPS: 3305
Stddev IOPS: 35
Max IOPS: 3379
Min IOPS: 3229
Average Latency(s): 0.0169412
Stddev Latency(s): 0.0272042
Max latency(s): 0.209305
Min latency(s): 0.00253389
Gerhard W. Recher
net4sec UG (haftungsbeschränkt)
Leitenweg 6
86929 Penzing
+49 171 4802507
Am 24.10.2017 um 16:31 schrieb Mark Adams:
> The last time I used ceph (admittedly a few years ago) I couldn't get it to
> go faster than about 30MB/s ... what sort of performance are you seeing
> with it now? I know this will be dependant on setup and workload, but if
> "all the best" hardware is used how fast can it go?
>
> Regards,
> Mark
>
> On 24 October 2017 at 15:28, Dietmar Maurer <dietmar at proxmox.com> wrote:
>
>>> The end goal for me here is a ZFS over iscsi setup with HA ZFS storage.
>> Do
>>> you have any other existing solution for this?
>> No - we always use ceph for such requirements ...
>>
> _______________________________________________
> pve-devel mailing list
> pve-devel at pve.proxmox.com
> https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
More information about the pve-devel
mailing list