On Thu, 2003-10-02 at 15:55, Adam Maloney wrote:
It was extremely nice to take the NNTP load off of our upstream links when we first set it up. As I understood it, they were not doing well on binary feeds towards the end there though.
I think they ended up filtering posts over a certain length over a year ago (?). They were approaching 45-50MBit/s, and when they implemented that filter they cut it back to about 30. Not exactly a full feed, but how much porn do you actually need? :)
I don't want to start speculating on certain issues, but I worked there between 4/00 and 4/01 as one of the engineers responsible for maintaining the uplink servers and other satellite doohickeys, so I can speak factually on certain events and paths we went down. Although Mike Donovan or Lisa Peoples would be able to explain much of this better than yours truly, I'll give it my best shot (as I remember it). As the Internet grew, NNTP traffic grew exponentially. Binary attachments were the bane of our existence, but... so long as we had the transponder throughput to accomodate our recipe of HTTP/NNTP/AV/etc, we avoided filtering as long as technically feasible. Unfortunately, it quickly became obvious that while NNTP was what was paying the bills (hypothetically... since too many ISP's were apparently too damn cheap to pay their bills), it was also choking the 45MB we could fit through the transponder. At one point in time, we were trying to push 250-260Gb/day across the transponder (roughly 22-30Mbps peak, IIRC). This left very little for our other "products". When it started to smother the rest, we were forced to start filtering on incomplete multi-part binaries. Some of our clients started bitching (some did from the beginning), as they would miss the occassional multi-part binary and blame Cidera. This was *not* any fault of ours, as we would push out everything we had. As a usenet peer, we were victim to incompletes just like anyone else (even with our excellent range of peer sources... thanks to M.D.). The only other type of "filtering" that might have occurred was throttling on the uplink. I have no doubt that things had changed drastically since the day I was laid off in April '01 (coincidentally, the day our SysEng staff went from 2 to 1). NNTP continued to increase, and likely always will. Folks like Donovan, Peoples, McGuire, Krokes, Humphrey and the rest did their damndest to provide a kick-ass product at a fraction of the cost of conventional terrestrial lines. I miss that place and the work we did with a serious passion. It was just one of those ideas and opportunities that doesn't come across very often, and I was damn lucky to be considered a [very] small part of it. *sigh* Cheers to the happy fun ball. -- Jason Dixon Former Systems Engineer Cidera, Inc.