Slow network performance to AWS

[gernthefish]

We have 3 on-prem servers that have stg pools in AWS/s3, and 1 EC2 server in AWS that also has s3 stg. Our local clients back up much faster to on-prem than they do to EC2 and the TSM cloud transfer from on-prem servers to s3 is about twice as fast as it is from the EC2 server to s3. All the servers are TSM v8.1.10.100 running on Linux RHEL6. All client backups are sent to the servers without any encryption or compression. Data is encrypted and deduped at the stg pool level. Client data is sent to the EC2 server via a VPN tunnel. We can ftp and scp through the tunnel at much higher speeds than we are achieving with TSM (about 5x faster). So it seems like there is something wonky with the EC2 server instance. Any thoughts?

 

[gernthefish]

Setting tcpwindowsize to 2048 seems to help, but client transfer rates are still much slower going to EC2 server than to on-prem servers. IBM is now suggesting we try tweeking txngroupmax (server) and txnbytelimit (client) settings.