Closed farcaller closed 4 years ago
Yeah, the s3 implementation is not tested against amazon s3, but against what ceph and riak provide. (riak was tested years ago, but I think that might still work). I think there's not much missing.
We use it (for tests) against FreeNAS S3. It works without many problems (see pull request) But it is not as fast as we hoped.
How fast is it? What are the parameters of your backy.cfg? What throughput can you reach with s3cmd or other tools when sending a number of 4MB blocks?
via s3 with
./s3cmd --no-ssl --host=10.10.1.200:9000 --host-bucket=test --access_key=nas2_s3_ednt_de --secret_key blablabla put /root/speedtest/* s3://test/
I get an average of 27MBytes/s
If I mount the drive via NFS or SMB and use rsync for copy, I get an average of 81MBytes/s.
But something is not correct, since the link is 10G and via iperf I can see that I have 9.8G which could result in 800MBytes/s. So I have to look what's wrong.
Since we use a ZFS RAID a higher speed as 81MBytes should be possible. Even when we use 'normal' SATA 6G drives. Btw. I used the same storage pool for S3 and the SMB/NFS share so the results should be comparable.
Well, actually that was not what I was asking for. I'll close the ticket as you seem to have internal performance issues not related to backy2.
version: 2.9.17
this targets amazon s3 in eu-central-1.