RustFS
Replies: 5 comments 12 replies
-
|
Thanks for your thorough work, it's really helpful for my selection! |
Beta Was this translation helpful? Give feedback.
-
|
Hello @xpnuo , Thanks for the detailed benchmark! The 4KiB results show a significant gap compared to our internal testing, and I'm not sure what caused this discrepancy. I’ll share our own 4KiB results with you for comparison. regarding large files: After receiving user feedback in July about poor performance, we have been continuously optimizing and have achieved great results. With our recent optimizations, our large file performance is now very close to MinIO. We will continue to optimize I/O speeds in the future. |
Beta Was this translation helpful? Give feedback.
-
|
Could you please change the log level to info and share the logs from the 512-thread crash? Thanks! We will fix this bug. |
Beta Was this translation helpful? Give feedback.
-
This rustfs issue is strange because rustfs-err.log is empty; there are no logs. I also can't authenticate when logging into the page via port 9001. Restarting rustfs with I can provide you with the logs from the warp load testing tool: root@seaweedfs-d-010057082043 seaweedFS]# ./warp mixed --host 10.57.82.44:9000,10.57.82.45:9000,10.57.82.46:9000,10.57.82.47:9000 --access-key admin --secret-key rustfsadmin --duration=1m --obj.size 256KiB --concurrent 512 Preparing: Clearing Prefix "warp-benchmark-bucket/". Deleted 1000 objects... λ warp: Io error: Remote disk operation timeout after 30s warp: upload error: One or more of the specified parts could not be found. The part may not have been uploaded, or the specified entity tag may not match the part's entity tag. 
|
Beta Was this translation helpful? Give feedback.
-
You can test the performance in cluster mode (if you've already tested it, you can also publish the results). It's very gratifying to see a product getting better and better. |
Beta Was this translation helpful? Give feedback.
-
This is error logs. Thank you very much. |
Beta Was this translation helpful? Give feedback.
-
|
This is error logs 
|
Beta Was this translation helpful? Give feedback.
-
This is my rustfs.log file. |
Beta Was this translation helpful? Give feedback.
-
|
Hello @xpnuo , Thank you for your feedback. The bug related to 512 threads with 1M has been resolved. You can find the fix here: 0927f93 Regarding your stress test results, 12 MiB/s seems incredibly low—we find that hard to believe. For reference, even on a minimal setup with 2 cores and 2GB RAM, we are able to reach 8 MiB/s. We will conduct further testing on our end. In the meantime, could you please let us know what your disk IOPS is? |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
this is my disk IOPS (When I run a 128-thread 4KiB object size): 
What command did you use? |
Beta Was this translation helpful? Give feedback.
-
|
Haha, what a coincidence! My main task in the first half of this year was to implement Milvus at the company, and I've been involved in their community activities. They also did some simple tests on RustFS and Minio recently; I can help you find their test report. |
Beta Was this translation helpful? Give feedback.
-
DDwarp put musl
warp put gnu
warp mixed gnuThe reason for the slow PUT speed is that READ consumes 23MB/s. 
|
Beta Was this translation helpful? Give feedback.
-
Test the performance of minio on the same disk for a better comparison. |
Beta Was this translation helpful? Give feedback.
-
|
Hi, thanks for sharing these benchmark results, very interesting work! Could you share more details on what is limiting the GET bandwidth? Thanks! |
Beta Was this translation helpful? Give feedback.
-
I confirmed with the IT team again that they only put one ssd disk on each machine. There seems to be a problem with the network bandwidth of the pressure test, so I submitted an issue to warp. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Hardware resources:
Test object version：
Deployment method: All are deployed in a cluster, running on node{1...4}
Benchmark commend: ./warp mixed --host node1,node2,node3,node4 --access-key --secret-key --duration=1m --obj.size <> --concurrent <>
RustFS GET
RustFS PUT
SeaweedFS GET
SeaweedFS PUT
MinIO GET
MinIO PUT
Summarize
Based on the results:
Small files: seaweedfs > minio > rustfs
Large files: minio > rustfs > seaweedfs
It's possible that the rustfs cluster mode is not yet mature enough, and it may crash during load testing (almost certainly reproducible) with 512 threads and a file size of 1 MiB.
I'm really looking forward to the development of RustFS. Its architecture is almost identical to Minio (which later replaced Minio and provided a great solution—and its performance is comparable to Minio for large files).
Beta Was this translation helpful? Give feedback.
All reactions