3 ms·
Seems easy to find: https://ceph.io/en/news/blog/2024/ceph-a-journey-to-1tibps/ https://ceph.io/en/news/blog/2024/ceph-a-journey-to-1tibps/
by pat2man 2y ago
Seems easy to find: https://ceph.io/en/news/blog/2024/ceph-a-journey-to-1tibps/ https://ceph.io/en/news/blog/2024/ceph-a-journey-to-1tibps/
- menaerus 2y ago3FS: 180 nodes, 2x200Gbps InfiniBand and 16x 14TiB NVMe SSDs per node, ~500 clients, 6.6 TiB/s of read throughput with training jobs workload Ceph: 68 nodes, 2x100Gbps Mellanox and 10x 14TiB NVMe SSDs per node, 504 clients, 1TiB/s of FIO random read workload
- ibotty 2y agoThe comparison is a little pears to apple. Similar nutritions but different enough to not draw conclusions. The hardware in the Ceph test is only capable of max 1.7TiB/s traffic (optimally without any overhead whatsoever). I also assume that the batch size (block size) is different enough that this alone would make a big difference.
- menaerus 2y agoEven if we take different hardware into account we can readjust for measured vs theoretical throughput. Ceph cluster achieves 1 TiB/s / 1.7 TiB/s = 0.58% of theoretical throughput. 3FS cluster achieves 6.6 TiB/s / 9 TiB/s = 0.73% of theoretical throughput.
- ibotty 2y agoThat difference is still pronounced, yes. But the workload is so different. Training AI is hardly random read. Still not a comparison which should lead you to any conclusions.