I have a Xen VPS with clustered.net running Ubuntu Linux 8.04 on a 2.6.18 x86_64 kernel. Apache 2.2.8 is serving a large test file. The VPS does not have load issues.
The measured throughput of wget from different IPs is as follows:
Peered ja.net: ~70Mbps
ADSL IP .17, Ubuntu 9.04: ~0.5Mbps (starts/stops a lot) Vista: ~0.5Mbps (starts/stops a lot)
ADSL IP .18, Ubuntu 9.04: 4.7Mbps (uplink max)
ADSL IP .19, Ubuntu 9.04: ~0.5Mbps (starts/stops a lot)
ADSL IP .20, Voyage Linux: 4.7Mbps (uplink max)
ADSL IP .21, Mac OS X 10.5: 4.7Mbps (uplink max)
Each of the ADSL IPs is on the same subnet and routed through the same switch and modem, assigned to different client machines. The two clients experiencing poor performance have a Tigon3 and Intel PRO/1000 NIC respectively. I tried an on-board RTL NIC in one as well, to no avail. Both clients have no problem receiving 4.7Mbps from any other peered mirrors (or > 250Mbps from LAN network mounts) and I have confirmed that all IPs use the same inbound and outbound routes to the VPS.
Here's the kicker: if I swap the IPs of any pair of good/bad throughput clients, by reconfiguring the LAN DHCP server, their throughputs swap as well. Neither the switch nor the modem runs a firewall and I temporarily disabled the VPS's firewall to no effect. Throughputs to all other tested sites remains high from any IP.
What on earth could be going on?
Edit: To be clear, all of the IPs are publicly routable. There is no NAT configuration in place.
Last edited by Jay Cornwall; 08-28-2009 at 07:10 PM.