Configuring bridged networking with vmware server 1.0.1.29996-r5 on dual amd64 machine with 2.6.16-gentoo-r9 kernel, bridged networking is incredably slow. http://www.vmware.com/community/message.jspa?messageID=403721 Reproducible: Always Steps to Reproduce: 1. Emerge vmware-server 2. Configure a virtual machine with a bridged ethernet adapter. 3. send traffic from host machine to guest machine, v. low bandwidth. some people report as low as 6k/bit sec. I got a little more, about 40k/sec. Though, my adapter is 1gb. Actual Results: Slow network traffic. Expected Results: Transfer network traffic approaching the speed of the network adapter. The site given above reports the following fix: "Try ethtool -K eth0 tx off ethtool -K eth0 sg off ethtool -K eth0 tso off After that bridged networking performance should be a bit better." However disabling protocol offload from the ethernet adapter is a poor solution as it can slow network traffic in general as work offloaded to the adapter is returned to the kernel.
Sorry but I fail to see what kind of solution do you expect from us.
Ok, fair point. I suppose that if I knew the solution I’d have posted it. I have posted a URL to a page that contains a proposed workaround others have found to be useful. After I emerge the vmware-server package (incl. other dependant packages) and after I install a guest OS in VMware, the host <-> guest network traffic is painfully slow. I regard this as a problem (bug/error), as do others. Using VMware on other systems (mainly windows) I've found that this type of network communication occurs at quite a high speed (approx the speed of the network interface). Do you propose bridged networking should have a transfer speed of kilobits, in which case are you suggesting that all other vmware systems on the planet have host to guest network traffic crippled in bridged mode? Or, alternativly some element in the VMware epackage distributed by the gentoo portage system interfaces poorly with network adapter drivers/kernel/? when "protocol offloading" to co-prosess incoming and outgoing network packets is used resulting in poor performance that can be overcome by correction...? Kilobits = not so good. megabits = good. Gigabits = much better.
I understand that the performance sucks, but its VMWare that needs to fix this, we pretty much can't. If the ethtool workaround works for you, then configure baselayout accordingly; we won't be messing with users' network configuration. There's a new version out, which should be in the tree shortly; if it's not fixed there, you'll need to bug upstream.
