You can't split a particular IP = connection=20 between two links, but can instead only determine which link a=20 particular connection will occur on. Given this, it sounds like = you=20 want to have some way to detect that Link A is already saturated and then s= end=20 all further connections to Link B until Link A is no longer=20 saturated.

Maybe someone can tell you how t= o do that=20 if that's really what you want to do (others here know far more about this = than=20 me), but my guess is you really don't want to do that. With the= =20 huge bandwidth disparity between the two links, route cacheing, a= nd=20 the inability know how much bandwidth any particular conneciton will= =20 consume, I think you'd end up with a giant mess... those people with connec= tions=20 unlucky enough to end up on Link A would probably be very unhappy people=20 indeed.

Generally speaking I think it wo= uld make=20 more sense to put all traffic over Link B, and then use Link A only for=20 emergencies. Maybe route the most critical traffic over Link A if you= =20 really want to feel like its being utilized as something other than a pure= =20 backup, but personally I wouldn't even do that.

Just because Link A is more reli= able and=20 more expensive doesn't mean it makes sense to use it as your primary=20 conduit. With Link B having eight times the bandwidth, it=20 seems the obvious choice as the primary. Use it, and keep the us= ers=20 happy most of the time (instead of making them miserable most of the= =20 time). On the rare ocassions it goes down, use bandwidth shaping to m= ake=20 sure the highest priority traffic gets access to Link A first.

In all the time I've used DSL, I= 've=20 had severe outages twice for reasons other than standard=20 maintenance. In both cases (in two separate locations), the cause was= the=20 ILEC phone company mistakingly dropping the wire pair while doing other wor= k=20 (freakin took over a week in each case to get my connectivity back!!). = ;=20 This sort of thing could just as easily happen with a leased line though, s= o I'm=20 not really sure I buy that the leased line is really more reliable than DSL= line=20 from a high quality ISP. Although maybe a particular SLA makes i= t so=20 in some legal sense since you can then sue someone. Personally, if yo= ur=20 leased line really costs more than the DSL, I'd get rid of it and get a 2nd= DSL=20 line from another provider and use that as your backup instead.

Anyway, I guess my main point is= that the=20 high cost of your leased line might be clouding your thinking on this. = ; I=20 wouldn't let the comparitive cost be your guiding light here. Go= with=20 what makes sense from a technology perspective, and don't guilt yourself in= to=20 trying to get full utilization out of the slow link just because it costs= =20 more.

----- Original Message -----

From:=20 Kenneth Kalmer

To: lartc

Sent: Saturday, April 23, 2005 4:3= 4=20 PM

Subject: [LARTC] Spill over

List

I need some help, advice or just a starting po= int=20 on the following situation:

Link A - 64kbps leased line
Link B = -=20 512kbps ADSL line

Is it possible to have Link A saturated constant= ly=20 and have the excess traffic "spill over" onto Link B? I know it= 's possible to=20 have packets sent down links in a round-robin fashion and I've read in th= e=20 howto on load sharing over multiple interfaces (http://lartc.org/howto/lartc.loadshare.html<= /a> ),=20 but I do not have control over the termination of the link at the ISP's (= two=20 different one as well). Also note that splitting different protocols over= each=20 of these links are not possible in our case.

Reason being, Link A = is a=20 more reliable and more expensive link, so I need to over-use it's capacit= y if=20 it we're, and use the cheaper ADSL (link B) offering to keep al services= =20 running when the leased line (A) is saturated.

Any tips, suggestio= ns=20 and comments would be welcomed.

Regards

--

Kenneth= =20 Kalmer
kenneth.kalmer@gmai= l.com
http://opensourcery.blogspot.com=20

_______________________________________________
LARTC mailing= =20 list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

_______________________________________________
LARTC mailing listLARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

--

Kenneth Kalmer
kenneth.kalmer@gmail.com
http://opensourcery.blogspot.com ------=_Part_5360_21792979.1114415300859-- --===============0919614172== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline _______________________________________________ LARTC mailing list LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc --===============0919614172==-- From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Taylor, Grant" Date: Mon, 25 Apr 2005 15:43:19 +0000 Subject: Re: [LARTC] Spill over Message-Id: <426D1017.2030109@riverviewtech.net> List-Id: References: In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: lartc@vger.kernel.org Kenneth Kalmer wrote: > Taylor & Chris (and the list) > > The arguments behind my choice here is cost driven, the 64kbps line is a > fixed monthly rate for unlimited use, the 512kbps line costs us roughly > ZAR250 per 3GB of usage. This can get quite expensive as the lines in > question is for a college and we all know what students do to bandwidth :) > > Taken the amount we pay every month for the 64kbps line it's more > economical to over utilize the link as a primary connection than to have > it lying around as a backup. South Africa and data connections don't go > well in the same sentence... > > As Chris suggested, I need something that can detect when Link A is > saturated and then redirect the traffic over Link B until there is > available bandwidth on Link A again. The rate limit trick of Taylor > might work once I get to understand the usage patterns of these > students. But for at least the first 3 months I won't have proper data > at my disposal. > > Thanks for your replies! Seeing as you are trying to avoid excessive bandwidth charges you find you self in a rather interesting position. Aside form what ever you end up using to control what route your traffic goes out you will ultimately want to have a fairly tight set up on what type of traffic is allowed. I have a client here in my town that managed to saturate an ADSL connection in the first 72 hours that we had the network in a testing state before we even went live with it. Ultimately we had to keep the traffic under 20 GB worth every month or they incurred extra charges. To this end I ended up setting up the router to only allow outbound traffic if it was destined to the following ports: (20) 21, 22, 23, 25, 53, 80, 110, 119, 143, and 443. I told the client that we would run with that set up until they told me that they needed another port opened for any given reason. I took the mentality of everything else will be shut down unless you can give me a reasonable request that is backed by someone with political authority (House monitor, Board Members, etc) that says that it is ok for you to be using this type of traffic. After I did this things have been GREAT. We have not had any more problems save for the month that they did not pay the ISP. The real problem that I see in trying to keep one link saturated is that you can control the outbound traffic but you have no way to control the inbound traffic. Well there are ways that you can attempt to control the inbound traffic but they are more a responsive control method (which is less likely to succeed) than a proactive control method like you can do if you control the sending end of the connection. A really drastic idea that I have that might not be too hard to implement would be to establish a PPP multilink session out both connections to a server somewhere on the internet that has HIGH bandwidth (more than your aggregate bandwidth). What this will allow you to do is control what link the traffic will come in and out. All your traffic would appear to be coming from the IP of the server out on the net but I don't see that as a problem really. I think if you do this you can set up routing and traffic control to try to use the 64 kbps link as the primary link and role over to the 512 kbps ASDL link if the first is saturated. I'm not sure how to go about doing this as this would be either in side of the PPP set up or some other complex method. You cold have traffic split across both links too as you control both ends and can reassemble it the way that it needs to be before it is sent out to the world or in to your LAN. One really nasty what that I could think to make this work would be to set up a default gateway to the IP of the ISP side of the 64 kbps link with a metric of 1 and a default gateway to the IP of the ISP side of the 512 kbps link with a metric of 2. You would need to watch the rate of traffic flowing out the 64 kbps link and any traffic that would be over it you would need to reject with no route to host which would cause your router to choose the other default route out, in this case the 512 kbps link. One MAJOR draw back to this that I see is that you would need / want to flush your routing cache fairly often, at least once per minute to make sure that your router would not end up learning that the 64 kbps link had problems and thus start using the 512 kbps link as a de-fac-to standard by remembering that the 64 kbps link did not have a route to any specific host. Something else that you should consider is that the subnets directly on the other side of each respective link have the best route to them of that said link vs going out the other link and hopping around on the internet to get back in to the first links IPSs network. Why take the long way around the building around the back to get from one front corner to the other? This is the type of paradigm that you will be creating. If this is not an issue you can disregard this statement. Needless to say there are a lot of issues involved here. This is not just a simple here you go type of thing. Grant. . . . _______________________________________________ LARTC mailing list LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Chris Bennett" Date: Mon, 25 Apr 2005 21:56:26 +0000 Subject: Re: [LARTC] Spill over Message-Id: <001501c549e1$a0b59090$200fa8c0@DELTA> MIME-Version: 1 Content-Type: multipart/mixed; boundary="===============1876831682==" List-Id: References: In-Reply-To: To: lartc@vger.kernel.org This is a multi-part message in MIME format. --===============1876831682== Content-Type: multipart/alternative; boundary="----=_NextPart_000_0012_01C549B7.B7D130B0" This is a multi-part message in MIME format. ------=_NextPart_000_0012_01C549B7.B7D130B0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable A little googling tells me 250 ZAR ~ 42 USD. Is this correct? If so, = ouch.. that's pricey. 3GB (assuming B in this case is BYTE) comes out to about 9kbit / second = over a month, if I did my math correctly. Ouch again. Does the 3GB apply to the total of up and down traffic, or just down? = Because you can't control traffic coming to you very well. You can try = to control TCP traffic with policing, but UDP traffic does its own = thing. Not to mention jokers who decide to flood the link for the hell = of it. Given this new info, it sounds more like you shouldn't try to use the = 512kbit link at all unless the 64kbit link goes down. If you do try to = push "excess" traffic onto it, all that does is encourage the use of = applications that will consume the entire bandwidth available. If that = is really beyond your budget, it doesn't seem like something you'd want = to do. Better to set the expectations at 64kbit so the users don't get = the idea of tuning into Internet radio or something. In fact, if the = 64kbit link does go down, it could be a good idea to police the 512kbit = link down to 64kbit, just so the users don't jump for joy when the = 64kbit link goes down... (keeping in mind that policing is no guarantee = that you'll actually stay below 64kbit usage, especially if a lot of the = traffic is UDP). ----- Original Message -----=20 From: Kenneth Kalmer=20 To: Chris Bennett ; Taylor Grant=20 Cc: lartc=20 Sent: Monday, April 25, 2005 2:48 AM Subject: Re: [LARTC] Spill over Taylor & Chris (and the list) The arguments behind my choice here is cost driven, the 64kbps line is = a fixed monthly rate for unlimited use, the 512kbps line costs us = roughly ZAR250 per 3GB of usage. This can get quite expensive as the = lines in question is for a college and we all know what students do to = bandwidth :) Taken the amount we pay every month for the 64kbps line it's more = economical to over utilize the link as a primary connection than to have = it lying around as a backup. South Africa and data connections don't go = well in the same sentence... As Chris suggested, I need something that can detect when Link A is = saturated and then redirect the traffic over Link B until there is = available bandwidth on Link A again. The rate limit trick of Taylor = might work once I get to understand the usage patterns of these = students. But for at least the first 3 months I won't have proper data = at my disposal. Thanks for your replies! ------=_NextPart_000_0012_01C549B7.B7D130B0 Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable
A little googling tells me 250 = ZAR ~=20 42 USD. Is this correct? If so, ouch.. that's = pricey.

3GB (assuming B in this case is = BYTE) comes=20 out to about 9kbit / second over a month, if I did my math=20 correctly. Ouch again.

Does the 3GB apply to the total = of up and=20 down traffic, or just down? Because you can't control traffic = coming to=20 you very well. You can try to control TCP traffic with policing, = but UDP=20 traffic does its own thing. Not to mention jokers who decide to = flood the=20 link for the hell of it.

Given this new info, it sounds = more like=20 you shouldn't try to use the 512kbit link at all unless the 64kbit link = goes=20 down. If you do try to push "excess" traffic onto it, all = that does=20 is encourage the use of applications that will consume the entire = bandwidth=20 available. If that is really beyond your budget, it = doesn't seem=20 like something you'd want to do. Better to set the expectations at = 64kbit=20 so the users don't get the idea of tuning into Internet radio = or=20 something. In fact, if the 64kbit link does go down, it could be a = good=20 idea to police the 512kbit link down to 64kbit, just so the users = don't=20 jump for joy when the 64kbit link goes down... (keeping in mind that = policing is=20 no guarantee that you'll actually stay below 64kbit usage, especially if = a lot=20 of the traffic is UDP).

----- Original Message -----
From:=20 Kenneth Kalmer