Mailing List Archived Message #52

From: "Lewis G Rosenthal" <> Full Headers
Undecoded message
Subject: TAP driver killing TCP/IP bandwidth
Date: Wed, 09 Dec 2009 14:01:33 -0500
To: Virtualized eCS Users <>

Greetings, all. I wasn't sure whether to post this to eCS-Technical, but I guess most of the possibly affected users may indeed be here (or on the VBox forum)...

Prompted by a post on the list for Peter Moylan's FTP server concerning high CPU utilization on gigabit links, I started running some tests the other night, using netio 1.26. What was interesting to find was that my eCS 1.2R server is barely getting 60Mbps on a gig switch, which then prompted me to do some more checking...

Eliminating the hardware for a moment, I snapped my ThinkPad T43 straight into the switch, and tested from there. Amazingly, I got similar bandwidth results, this time using GenMAC, for the same B57 NIC as what is in the server - or at least a close cousin).

Switching to the native B57 driver, the results got even worse, down to the single digits.

I moved back to the GenMAC driver and started eliminating things from CONFIG.SYS to try to narrow down what was causing the bottleneck. IPX? No. IJFW? No. It turned out that the VBox TAP driver (PROT.OS2) was the culprit. Commenting that out of CONFIG.SYS brought my connection throughput back up to the smae level I am seeing between Linux boxes, around 850Mbps. (Note that for testing purposes, I had a 7' CAT-6 cable connected to the ThinkPad and a 15' CAT-5e cable connected to the Linux workstation, both going through a LinkSys SR2024 10/100/1000 unmanaged switch; I'll have to fashion myself a CAT-6 crossover to eliminate the switch from the equation.)

Has anyone else seen this type of performance hit cuased by the TAP driver?


Lewis G Rosenthal, CNA, CLP, CLE
Rosenthal & Rosenthal, LLC      
Need a managed Wi-Fi hotspot?      
Secure, stable, operating system

Subscribe: Feed, Digest, Index.
Mail to ListMaster