When FileCatalyst first introduced congestion control in 2006 (or v1.5 of the software), we decided to have the protocol “play nice” on networks.
To a network administrator, this meant that we never oversaturated their network when they ran their first UDP transfer. It also meant that with HotFolder's default settings, transfers would begin at 384 Kbps and ramp up to transfer at 1.5 Mbps.
Not exactly rocket speed.
Canadian nice on the network was accomplished by setting values artificially low during installation:
- Congestion control settings were set to slow down aggressively upon hitting congestion and speed up slowly when congestion cleared up (the “aggression level” was set to 2 out of 10). If TCP traffic was detected, our software dramatically slowed down and let it through.
- Congestion control start rate was set to a low value (384 Kbps), so most customers would never encounter an adverse network burst when starting a transfer for the first time.
- HotFolder transfer rate was limited to a maximum of 1544 Kbps (1.5 Mbps), ensuring that we would never over saturate a link upon first installation.
As it turns out, the "safe" settings aren't what the administrators want as an out of the box experience. They want to give their users fast transfers without any tinkering. Not even a single setting.
We hear you, loud and clear!
In FileCatalyst v3.1, we decided to bump our default settings to give you more performance right from the start.
- HotFolder bandwidth is now set to a maximum of 1 Gbps (up from 1544 Kbps).
- Applet bandwidth is now set to 100 Mbps (up from 10Mbps)
- Congestion Control Aggression is now set to level 5 out of 10 (competitive with TCP traffic) instead of the old default 2 (nice to other traffic).
- UDP Start Rate assumes you have at least a 1Mbps connection (instead of 384 Kbps)
- Packet Loss Strategy is the new default algorithm (instead of RTT-only), which allows for quicker speedup and slowdown when congestion is detected on the line (see here for a brief explaination: http://www.filecatalyst.com/congestion-control-in-filecatalyst)
These changes are not a complete reversal of our nice-guy settings (we could have set aggression to 9, and watch TCP traffic get shoved out of the way). Nor do they mean that you will never choose to make changes to your settings; over time, a well-tuned deployment will maximize transfer efficiency. But they do help meet new customers' expectations for our software upon first installation, and set a baseline scenario that's more in keeping with today's file transfer needs.