Dev VM Network Slowdown — Checkpoint
Issue: Severe sustained-transfer throughput collapse between Andrew's CentOS 6 Dev VM and Live server (forum.drugs-and-users.org, 128.199.200.202, DigitalOcean droplet). Dev system currently unusable for large rsync of /var/www/html/.
Symptom
- scp/rsync transfers start fast (hundreds of KB/s to low Mb/s) then throughput collapses/decays the longer they run — both push and pull directions.
- One-off speedtest-cli / curl tests report healthy throughput (~40 Mb/s) — only sustained transfers are affected.
Tested and ruled out
- Remote disk I/O (iostat/dd) — clean
- Remote CPU/load (top/uptime) — low
- Local CPU/steal time (vmstat) — steal=0
- Local NIC errors (ip -s link) — flat error count under load
- TCP retransmits/cwnd shrinkage (ss -ti) — retrans:0/1, stable cwnd
- Duplex/ethtool — vmxnet-style driver, no useful data
- NIC offload settings (ethtool -k) — mostly already off; GRO on but untested as fix
- tc qdisc shaping on remote — none, default pfifo_fast
- iptables rate limits on remote — only ICMP/new-connection limits, not bandwidth
- CSF — not installed on remote
- fail2ban — installed, ssh-iptables jail active, Andrew's IP not banned
- IPv6 fallback — forced -4, no change
- /etc/hosts override — none, resolves correctly to real public IP
- NAT vs Bridged on Dev VM's network adapter — switched to Bridged as a test, result was WORSE, reverted back to NAT
- Disk resize (30GB→60GB) on Live droplet — confirmed clean: partition table and filesystem correctly matched at 60G (df/lsblk/proc-partitions all agree), usage now 48% (27G/60G) vs earlier 83%
Not yet tried
- iperf3 raw throughput test (isolates from SSH/scp/rsync overhead entirely)
- Checking DigitalOcean-side bandwidth policing/plan limits on the droplet
- Checking DO droplet event history for a reboot/migration coinciding with the disk resize
Plan
- Andrew rebooting the Windows host machine that hosts the Dev VM to see if that clears it.