makssent commented on issue #36972:
URL: 
https://github.com/apache/shardingsphere/issues/36972#issuecomment-3600546834

   I checked the network between the Proxy and all cluster nodes. I ran the 
following tests on each node:
   
   ```
   iperf3 -c NODEX -t 10
   iperf3 -c NODEX -P 4 -t 10
   iperf3 -c NODEX -R -t 10
   ```
   
   Results without Proxy running:
   
   | Node  | Single         | Multi (P=4)    | Reverse        |
   |-------|----------------|----------------|----------------|
   | NODE1 | ~23.5 Gbit/s   | ~29.2 Gbit/s   | ~22.9 Gbit/s   |
   | NODE2 | ~21.8 Gbit/s   | ~28.4 Gbit/s   | ~24.5 Gbit/s   |
   | NODE3 | ~23.5 Gbit/s   | ~21.8 Gbit/s   | ~23.6 Gbit/s   |
   | NODE4 | ~25.0 Gbit/s   | ~29.7 Gbit/s   | ~23.3 Gbit/s   |
   | NODE5 | ~23–27 Gbit/s  | ~29–30 Gbit/s  | ~26.4 Gbit/s   |
   
   The network works stably in the 22–30 Gbit/s range, without drops or 
instability. With Proxy running (but without OLTP load), throughput remains at 
the same level — around 22–24 Gbit/s.
   
   When OLTP tests (200 threads) are running, iperf3 and Proxy begin to compete 
for CPU and the network stack. Because of this, iperf3 speed initially stays 
around ~20–23 Gbit/s, but then drops:
   
   ```
   743 MBytes   6.12 Gbits/sec
   680 MBytes   5.79 Gbits/sec
   602 MBytes   4.99 Gbits/sec
   580 MBytes   4.99 Gbits/sec
   ...
   208 MBytes   1.65 Gbits/sec
   165 MBytes   1.27 Gbits/sec
   137 MBytes   1.15 Gbits/sec
   125 MBytes   1.12 Gbits/sec
   ```
   As soon as the load decreases, the speed returns to the normal level:
   
   ```
   [  5]  57.00-58.00  sec  2.53 GBytes  21.7 Gbits/sec
   [  5]  58.00-59.00  sec  2.75 GBytes  23.7 Gbits/sec
   [  5]  59.00-60.00  sec  2.67 GBytes  22.9 Gbits/sec
   - - - - - - - - - - - - - - - - - - - - - - - - -
   [ ID] Interval           Transfer     Bitrate
   [  5]   0.00-60.00  sec  86.5 GBytes  12.4 Gbits/sec                  sender
   [  5]   0.00-60.00  sec  86.5 GBytes  12.4 Gbits/sec                  
receiver
   ```
   
   Even the minimal speed (1–2 Gbit/s) is still higher than the actual SQL 
traffic of the Proxy. The network between Proxy and nodes is not a bottleneck. 
There is more than enough throughput for all OLTP tests, and no network 
limitations are observed.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to