Have been doing a network analysis contract and it’s brought up a couple of interesting misconceptions. Their office has a cable line (50Mbps down and 4Mbps up) for WAN1. It also has a backup ADSL line (8Mbps down and 2Mbps up) for WAN2. There were constant problems with speed and connection on WAN1, and their idea had been to upgrade the line to 200Mbps (6Mbps up).
This seems pretty reasonable at first – a higher download speed should equate to faster browsing. ISPs sell fast packages but misuse terms quite frequently. This results in customers making understandable assumptions that don’t fix the problem. In this case, the office thought that a faster download speed would mean faster downloads. They didn’t figure, however, for the effect of the upload speed.
Finding the problem
Having booted the majority of low-bandwidth clients onto WAN2, the culprit on WAN1 was obvious. There was a single machine, using Skype, that was running a pretty high upload rate (see the blue line below). The minute I kicked the client, the upload bandwidth use dropped to nearly nothing. As a result download speeds started increasing (see the green lines below). There wasn’t a lot of download traffic but the flatline is pretty obvious.
Skype seemed to be the main culprit, but the client was also logged in to a lot of sites and services (Facebook and iTunes, for e.g.) and killing all of them completely freed the upload rate.
TCP/IP traffic handshakes
The two rates are related because TCP/IP traffic works both ways. Each packet of data that gets sent is stamped and verified. If it’s rejected or communication drops, the packet gets resent. The problem here was that the upload rate was running at its maximum, so there wasn’t enough of it left to either initiate new connections or properly maintain current ones. There was all this download rate available but not enough upload rate to unleash it.
With that in mind the upgrade to 200Mbps didn’t seem that effective with a gain of only 2Mbps up, despite another 150Mbps down, and if one machine can saturate a 4Mbps up then I can’t see it helping. So initially we’ve stuck in a rate limit cap at .5 Mbps for each machine, which doesn’t help the one client but at least makes it a bit fairer on the rest of the office, and we’re going to start configuring the apps and routers individually (luckily it’s a small office).