how-to block ads
|
larek
join:2004-03-25
Hancock, MI
1 edit | [DSL] Modem Poisoning
Provider: ATTyahoo
Connection: 4x6mb Dsl connections
Modems:
Speed streem 5100 "B"
Speed streem 5100 "B"
Speed streem 4100 B
Speed streem 4100 B
Router: XINCOM X16-R
I have a problem that has plague our house for years. We call it modem poisoning. It is where a Modem appears to be functioning (provides link, DHCP services, and dsl light) but all traffic ceases to get anywhere.
This occurs on all of my modems, on any dsl line. The only fix is hard powering off of the affected modem.
Poisoning makes our 16-Wan load balancer useless, as the router cannot tell that a modem is Poisoned since other than not passing network traffic on it appears OK, and the router will send most or all of the traffic to the Poisoned one, to to its low/nonexistent band with use.
We've found that the Poisoned effect now be recreated by opening large number of active connections are opened in a short period of time. U-Torrent is the easiest way to accomplish this, set active connections to 200+ and connect to a popular torrent. A Lower Max connections value can cause poisoning too if they happen fast enough.
All our modems are set to use PPPoE internally. I herd a long time ago that the Resources (RAM) on-board a DSL modem (and cheap routers) are quite limited and that you can over-fill their internal buffers by doing things like attempting things like too many connections. Does this ring any bells with anyone?
Has anyone else had issues like this, and or found a soultion?
Would a third device between, each modem and the router to handel the PPPoE, that has say Gigs of ram fix this issue? | |
|  larek
join:2004-03-25
Hancock, MI
2 edits | Re: [DSL] Modem Poisoning Common Solutions I Hear
Limit your max connections:
Its true if I limit my Max Connections down to 24, a single modem doesn't get seem to get poisoned. However that is program specific, and there are 6 college students living in my house. So each would thus have to limit their total max connections to 4 (for all their computers and programs). It would not be 24x4modems/6users since with the load balancer you cannot guaranty that the active connections would be spread evenly. Remember poisoning one modem brings down the whole house when using the load balancer. No one wants to limit themselves to 4 connections max. I don't really want to limit my max connections at all.
Change your Loadbancer Config:
True if I change the balancer to say Round Robbin, it does keep some Internet alive. But its really poor QOS. Even web surfing with 1/4 of the traffic just lost (require active retrying by the user) is very annoying.
Is your cabling bad?
No, I had good a dedicated cat5 wiring infrastructure. But due to (suggestions) it has been replaced with dedicated Shielded Cat6, Each modem has its own discrete shielded cable, direct to the Tel-co (ATT) box on the exterior of the House.
Don't load balance:
Use one modem per person.. That's what were currently doing, but is less than optimal and 4 modem's don't enveny split 6 ways. And were already paying $200/month so adding more modems+lines isn't viable.
AND when the load balancer works the speed 12mbish is amazing so we really so want to use it. | |
| 
d_l
Barsoom
Premium,MVM
join:2002-12-08
Reno, NV
| Those are 5100b modems right? Because the 5100a modems won't make the connection.
I thought the maximum number of connections one of those modems could handle was higher, say maybe in the 40-60 range, but this is from memory so maybe wrong. I do know that if they aren't making the PPPoE connection then their connection "limit" is raised because the modem memory devoted to PPPoE is used for sessions instead.
One reason those modems are so connection limited is that their software is using some (a lot?) of the internal memory for ACK prioritization. Other modem models such as the notorious 2210 modems or the 2wires or older bridged modems don't prioritize ACKs and so you can have many connections open, but your download is choked to upload speed levels when doing intensive uploading.
Last year some experiments were were carried out to flash the AT&T version 4100 with generic 4100 software (actually it was an Australian ISP version of generic software). The connection limit was raised to the thousands before the modem would crash. ACK prioritization was eliminated though. | |
| | larek
join:2004-03-25
Hancock, MI
| Re: [DSL] Modem Poisoning 5100(b/a)? not absolutely sure, I'll check at lunch
The connection in the 40-60 range I don't know. 60-80 do die but it its sort of random. 200+ is guaranteed a quick death thats why I mentioned it. 24 is fairly stable.
I haven't done any testing with ACK prioritization on/off since I thought that PPPoE was the major resource consumer. I will start now test it too.
If the speedstreems are connection limited from ACK prioritization (and the limited resources that it consumes) Do ADSL modem cards (PCI?) have better performance since they can use the full system resources? Can the use the full pc system resources?
So if I offload the PPPoE and ACK prioritization, off the modem to a third box "PPPoE Server", and have the Modem in bridge mode, my problem might just go away? | |
| | | larek
join:2004-03-25
Hancock, MI
2 edits | Re: [DSL] Modem Poisoning Nope I checked again they say just "5100", no A or B seen on the back.
What is meant by the term Generic Firmware? I assuming you don't mean the original factory Firmware. I read many posts that mention it in passing. Do you really mean 3rd Party Firmware, or non-provider branded firmware.
I'm wondering if ACK p could be turned off via the CLI command?
•Note: I'm asking about Firmware, not for Firmware
| |
| | | | 
wayjac
Premium,MVM
join:2001-12-22
Indy
 ·AT&T Midwest
| Re: [DSL] Modem Poisoning said by larek  :Nope I checked again they say just "5100", no A or B seen on the back. There are pictures of the 5100a, 5100b, 4100 and 4100b in this faq
AT&T HSI / Ameritech Modem Museum
said by larek :What is meant by the term Generic Firmware? The speedstream 4100 and 4200 router gui are almost identical
Below is a image of the speedstream 4200 router gui
| |
| | | | | larek
join:2004-03-25
Hancock, MI | Re: [DSL] Modem Poisoning Ok so mine are the the 5 LED "b" version. They are from 04' | |
| | |
wayjac
Premium,MVM
join:2001-12-22
Indy
·AT&T Midwest
| said by larek :So if I offload the PPPoE and ACK prioritization, off the modem to a third box "PPPoE Server", and have the Modem in bridge mode, my problem might just go away? I think the problem is the nat table size more than ACK prioritization, can the load balancer be configured to do the PPPoE
The load balancer's nat table may be able to handle the many connections | |
| | | |
d_l
Barsoom
Premium,MVM
join:2002-12-08
Reno, NV | Re: [DSL] Modem Poisoning Yes, normally these modems simply crash or reboot or drop sync when they have too many connections. | |
|
d_l
Barsoom
Premium,MVM
join:2002-12-08
Reno, NV
| ACK prioritization is just there in software in the 5100b, 4100, and 4100b modems. You can't disable it other than flashing to a generic modem software. I only know of versions for the 4100 modem, not the 4100b or 5100b.
The PPPoE client, on the other hand, can be moved to a router. This will increase the connection limitation on any one modem by maybe as much as double that as when the modem is making the PPPoE connection.
Whether that will solve your network wide connection problem, it is hard to say. If you could effectively double the number of connections each modem handle, I don't think that will be enough if the load balancer isn't balancing them across the modems.
If you want a lot of connections per user you will have to do that rather than balancing by bandwidth or you will have to have modems that allow a much greater number of connections, e.g. non-ACK prioritizing modems. | |
| |
d_l
Barsoom
Premium,MVM
join:2002-12-08
Reno, NV
| The 5100b has 5 LEDs vs the 4 of the 5100a.
By "generic" firmware, I mean a version of firmware that allows access to the router functionality with the GUI and would basically conform to the user manuals that Siemens used to publish. Each ISP has changed various minor features within the software, but AFAIK, no other ISP has changed the GUI to the same extent as AT&T(SBC). There are no third party versions of the software.
Can ACK prioritization be disabled in the CLI? Good question. I couldn't find it. There is a TCP/IP QoS feature listed in the generic modem manual, but I didn't think that to be the same as ACK prioritization. There are CLI commands for that QoS though.
As of about a year and a half ago, there were Australian and Brazilian ISP (Portuguese/English) versions of the software available to flash the SBC 4100 to a more generic 4100/4200 modem and it has been done. I think there are various softwares available that would flash an SBC version to a more generic version of the 5100/5200 modem, but I've not heard of anyone trying this. The 4100b might be a special case. There seem to have been some design revisions within the 4100b (aka 4101) and there may not be a generic software version available that would work on it. IOW, the AT&T 4100b hardware may be unique to AT&T. | |
| NormanS
Premium,MVM
join:2001-02-14
San Jose, CA
·Pacific Bell - SBC
| said by larek :Has anyone else had issues like this, and or found a soultion? No. I have nothing like what you describe set up. However, I found this link:
»www.xincom.com/products/X16-R/specs.php
According to the specs, that Xincom X16-R is capable of multiple PPPoE sessions. I wonder if bridging all of the modems, and configuring the Xincom to run each PPPoE session would help?
--
Norman
~Oh Lord, why have you come
~To Konnyu, with the Lion and the Drum | |
| | larek
join:2004-03-25
Hancock, MI
1 edit | Re: [DSL] Modem Poisoning said by wayjac : The load balancer's nat table may be able to handle the many connections The router, [other than it's inability to detect a poisoned modem] is great. Its only locked up once in the years we had it.
The lockup is on the modem end, and they are just plain modems, not modem+routers, so I don't believe they have a nat.
said by NormanS :I wonder if bridging all of the modems, and configuring the Xincom to run each PPPoE session would help? That were the original idea came from, it can do PPPoE for all 8 WAN connections, but a single poisoned modem, brings down the router. We have a regualr weekly online game, and need to be able to switch from the router to a dedicated modem fast. (until the system is proven at least)
That's the one of the major reason to off load that PPPoE to a "Box" is that the "Box" can sit between both the router and modem, or PC and modem.
b_l suggested that ACK prioritization may be a bigger resource consumer than PPPoE, and that Xircom can't do that so a "box" is needed anyway to test that theory. | |
| | |
wayjac
Premium,MVM
join:2001-12-22
Indy
·AT&T Midwest
| Re: [DSL] Modem Poisoning said by larek :The lockup is on the modem end, and they are just plain modems, not modem+routers, so I don't believe they have a nat. Believe it or not The modems have nat as a option when the modem is configured to do the PPPoE
You can configure the modem so that it is a bridge
said by larek :That's the one of the major reason to off load that PPPoE to a "Box" is that the "Box" can sit between both the router and modem, or PC and modem. It will be hard to find a "box" that will just do the PPPoE only. Hence the suggestions that you bridge the modems and let the Xincom handle all the PPPoE connections
When the speedstreams are set to do the PPPoE you also have the option to use a private ip address for the device connected to its lan port, the default is to use the public ip address.
Which option are you using on the modems? | |
| | | |
|
