dslreports logo
site
 
    All Forums Hot Topics Gallery
spc

spacer




how-to block ads


Search Topic:
uniqs
27
share rss forum feed

dynodb
Premium,VIP
join:2004-04-21
Minneapolis, MN
reply to knightmb

Re: Good step

said by knightmb:

There is no reason to worry about a customer using the pipe 24/7 really. As one who runs his own ISP, I've found that all the myths that the big players say about bandwidth hogging and such are just false. It really just comes down to knowing how to configure your routers, gateways, and portals properly.
Please enlighten us as to how one might configure their router, gateway or portals to allow a DS3 or OC3 to pass more than 45M or 155M of data.

iansltx

join:2007-02-19
Austin, TX
kudos:2
Compression?

Seriously, some traffic is more important than others. You can do this based on flows and do it transparently...


knightmb
Everybody Lies

join:2003-12-01
Franklin, TN
reply to dynodb

A torrent state table
Click for full size
Torrent in action
said by dynodb:

said by knightmb:

There is no reason to worry about a customer using the pipe 24/7 really. As one who runs his own ISP, I've found that all the myths that the big players say about bandwidth hogging and such are just false. It really just comes down to knowing how to configure your routers, gateways, and portals properly.
Please enlighten us as to how one might configure their router, gateway or portals to allow a DS3 or OC3 to pass more than 45M or 155M of data.
Not enough space to fit it all here if you want technical details, but mainly it boils down to not even looking at the traffic in terms of how much bandwidth you have, but more in the terms of making sure traffic is shared evenly to begin with.

If you have a node that will max at 45 megabits/sec and it's not smart enough to know how to pipe in simultaneous connections into a group so that other traffic can pass it's time to look at some other hardware setups. Quite frankly, it is basically traffic shaping but without all the silly rules. If you have one pipe (use the 45 megabits/sec again) and 45 customers tied to this node; then one customer running BitTorrent should only be able to dominate the idle available bandwidth. If the customer is able to burn 45 megabits/sec and stall everyone else instead of the bandwidth being split down evenly among all connecting peers, then that's another example of not using some basic traffic shaping setup.

The biggest problem I see though is mainly the state tables being way too small. You can have a switched network where computer A and computer B doing a file transfer at full bandwidth doesn't halt everyone else in the office. The problem occurs for some devices where something like BitTorrent isn't churning out bandwidth as much as many simultaneous connections. So the gateway device that is awesome for speed when peers are just sending data to a few IP address chokes when one peer is just sending small chunks to hundreds or thousands of IP address at a time. The device just wasn't designed with the RAM needed to maintain a state table that big and thus just chokes out.

It's the same problem that all home routers have, great bandwidth for a few connections, but crash/choke they are asked to do small bandwidth to a lot of different locations. They all just run out of RAM because the state table is just too big to fit.

Your desktop PC suffer the same issue as it can only make so many connections before it hits a limit. The real key is making sure whatever setup you have is going to more powerful than anything a client (PC) could throw at it be it just raw data transfer or hundreds/thousands of connections going at once.

For me anyway, FreeBSD and Linux have basically solved the issue due to how much fine tuning they allow. A good PC with lots of RAM does circles around some of the most expensive commercial software/hardware out there.

Sure, you can buy your way into the multi-hundred thousand dollar Cisco equipment that would be on par with the setups I'm talking about, but I don't believe that even the richest ISP (Comcast, AT&T, etc.) are going to spend that kind of money to install one at every block to make sure John Doe and Jane Doe down the street can check e-mail while D00D1 down the block is sharing his torrents with as many clients as his PC/Connection can handle at once.

The next best thing you can do is build it yourself and spend all the money saved on keeping the pipe big for all your customers and not worry about which time of day and how much bandwidth percentage XYZ customer is using. We have customers that barely touch 200MB a month and others that doing 10GB a day of traffic. The only company policy we have is keep it fast, don't mess it unless it becomes a problem.

Hmm, this reply was a lot longer than I expected and barely touched on the technical side of things, so basically if you want to churn out bandwidth like this screenshot I have attached (just picked the top 100 from thepiratebay as a example, stealing is wrong, etc. yes I know, canceled it after the screenshot)
then FreeBSD gateway is your friend (and Linux too for a OS for the client). I had over 6,000 connections going, of which over 5,000 just for that torrent. So even with all the leechers, was still about to churn out some good bandwidth for it. That's running on the same pipe as 3 other customers who having their BT blazing 24/7 also, but their machines barely touch 500 connections due to probably defaults or limitations of their PC. In all that, other PCs still surf the web, play movies, etc. as though nothing is wrong.

If I can burn up this much simultaneous bandwidth while 3 others are doing the same and everything hums along just fine, I just shake my head when I hear hype about the one BT user who is able to take out his/her entire neighborhood with just a single PC. It's all about the setup.
--
Fight Insight Ready (Was NebuAD) and the like:
Click Here to pollute their data


lovswr

join:2001-09-15
Saint Petersburg, FL
reply to dynodb
Let me pick a nit. The SPE (SONET Payload envelope) in an Oc-3/STM-1 is 140M. There is 15M of overhead that the system uses for its own operation. Such things as the stuffing events, the payload pointer (which tells the system where YOUR 140M of data starts within each frame, & even a orderwire channel that is big enough for analog broadcast video.

So yes the line or wire speed is clocked at a little over one hundred & fifty million times per second, the actual throughput or amount of customer data is not quite that high.
--
lovswr = good hivswr = bad


knightmb
Everybody Lies

join:2003-12-01
Franklin, TN
said by lovswr:

Let me pick a nit. The SPE (SONET Payload envelope) in an Oc-3/STM-1 is 140M. There is 15M of overhead that the system uses for its own operation. Such things as the stuffing events, the payload pointer (which tells the system where YOUR 140M of data starts within each frame, & even a orderwire channel that is big enough for analog broadcast video.

So yes the line or wire speed is clocked at a little over one hundred & fifty million times per second, the actual throughput or amount of customer data is not quite that high.
Nit pick the nit pick I thought an OC-3 running STM-1 operated at 155 M/bs with 150 M/bs due to payload overhead? I'll have to dig out a manual for that one I guess.

The fiber networking protocol you refer to though is *usually* protocol neutral. It's job is just through the packets out (over-simplifying of course given routes, etc.) as fast as it can. Being that an atomic clock keeps the network in sync, it kind of sits high up on the hardware chart for me. I've never really seen one of those taken down by overload, but I know it does happen. As much as I would love to talk about overhead, payload, and pointers; that might cause a real eye-glazing effect at this news forum.

Basically, my only spill is that (and you probably know as well), that some of the problems we hear about with an ISP overload does have a solution that doesn't involve punishing one user for being a bandwidth hog or deciding that everyone should just be capped to XYZ gigabytes of bandwidth.

I know that no system is invincible to load. I could get about 30 clients doing about 30,000 connections each to max out the ISP setup due to the gateway/router devices having only a few gigs of RAM for state tables if I really wanted to push it to destruction.
--
Fight Insight Ready (Was NebuAD) and the like:
Click Here to pollute their data


88615298
Premium
join:2004-07-28
West Tenness

1 recommendation

reply to knightmb
said by knightmb:

Even though you tried to blur it out I can see you are illegally downloading Transformers 2. People like you is the EXACT reason why we have to put up with caps in the first place. Thanks. Is $5 too much for you to afford to rent it from the video store?