Tracing DSL delays

I have a client who has been successfully using DSL for their web
connection. Recently, I replaced their ancient server (a PII-233) with a
brand new, Dell super-duper box. This server is way overkill for them,
but they wanted something that they wouldn't have to worry about.

Coincidentally with this install, their DSL connection went to crap.
Their download speeds vary from 10K to 40K, but they should be seeing 58K
downloads. Web sites time out. DNS queries fail.

We eliminated the server as the cause. While it provides DNS and DHCP,
the local DNS resolution is quick, it's the off-site DNS that's slow.

What I'm looking for is some sort of tool that can long-term monitor
what's going on. I'd like to run maybe a 12 hour test throughout a
business day, to see if it varies with the time of day. My tests, run
after business hours, show normal speeds. Their tests, done DWH, show
slow speeds.

But I need something that I can take to the ISP and show them where the
problems lie. The ISP is very cooperative, but they're stumped as well.
They've replaced the DSL router, run their tests on line quality, and
everything checks out, except that my client has crappy web access....

I'm guessing it's some sort of intermittent failure or overload on the DSL
loop.... But I'd like some data to back it up.

So what can I use to collect useful data to show to the ISP?

--

Civil Engineer for hire - http://www.seiner.com/engineer.html

use munged address above to email me
SpamTrap (E-Mail Removed)

"Captain Dondo" <(E-Mail Removed)> wrote in message
news(E-Mail Removed) om...

> Coincidentally with this install, their DSL connection went to crap.
> Their download speeds vary from 10K to 40K, but they should be seeing 58K
> downloads. Web sites time out. DNS queries fail.

Have you disabled Explicit Congestion Notification (ECN) to see if that is
the issue?

Depending on your distro, it may be as simple as:

echo 0 > /proc/sys/net/ipv4/tcp_ecn

On Wed, 09 Mar 2005 09:35:44 -0800, Captain Dondo wrote:

> I have a client who has been successfully using DSL for their web
> connection. Recently, I replaced their ancient server (a PII-233) with a
> brand new, Dell super-duper box. This server is way overkill for them,
> but they wanted something that they wouldn't have to worry about.
>
> Coincidentally with this install, their DSL connection went to crap.
> Their download speeds vary from 10K to 40K, but they should be seeing 58K
> downloads. Web sites time out. DNS queries fail.
>
> We eliminated the server as the cause. While it provides DNS and DHCP,
> the local DNS resolution is quick, it's the off-site DNS that's slow.
>
> What I'm looking for is some sort of tool that can long-term monitor
> what's going on. I'd like to run maybe a 12 hour test throughout a
> business day, to see if it varies with the time of day. My tests, run
> after business hours, show normal speeds. Their tests, done DWH, show
> slow speeds.
>
> But I need something that I can take to the ISP and show them where the
> problems lie. The ISP is very cooperative, but they're stumped as well.
> They've replaced the DSL router, run their tests on line quality, and
> everything checks out, except that my client has crappy web access....
>
> I'm guessing it's some sort of intermittent failure or overload on the DSL
> loop.... But I'd like some data to back it up.
>
> So what can I use to collect useful data to show to the ISP?


If you ever figure out what exactly is happening and get the issue
resolved (no pun intended) PLEASE post back the info.

Meantime, to alleviate the situation - you need

options timeout:1

in you /etc/resolv.conf file. It seems that the issue has to do with
certain DNS servers not catching the first (or first few) Linux DNS
requests so the system has to time out one or more times before it
actually resolves.

On Wed, 09 Mar 2005 09:35:44 -0800, Captain Dondo wrote:

> I have a client who has been successfully using DSL for their web
> connection.

What kind of DSL? ADSL? SDSL? Web connection? Does that mean serving
web sites or web surfing?

> Coincidentally with this install, their DSL connection went to crap.
> Their download speeds vary from 10K to 40K, but they should be seeing 58K
> downloads. Web sites time out. DNS queries fail.

Web sites timing out sounds suspiciously like MTU issues. I'm assuming
your DSL line is using PPPoE. Which means your internal hosts need an MTU
of no higher than 1492, and it may need to be lower than that.

> We eliminated the server as the cause. While it provides DNS and DHCP,
> the local DNS resolution is quick, it's the off-site DNS that's slow.

What other services are you providing over this connection? If it's an
ADSL line, then you have to remember that your uplink is the limiting
factor. If a user were to send a large email out over the ADSL line, the
uplink could easily be saturated and prevent anyone else from utilizing
even a fraction of the downlink.

Saturating your uplink can also cause latencies in excess of 5000ms.

> What I'm looking for is some sort of tool that can long-term monitor
> what's going on.

If you're running GNU/Linux on the router, use tcpdump to sniff a specific
connection that you know is being problematic.

> business day, to see if it varies with the time of day. My tests, run
> after business hours, show normal speeds. Their tests, done DWH, show
> slow speeds.

This sounds traffic related.

> But I need something that I can take to the ISP and show them where the
> problems lie.

Ultimately DSL is a consumer class service. The ISP isn't going to sink
the same resources behind a $100/month service that they would behind a
$1000/month service.

> So what can I use to collect useful data to show to the ISP?

Again, if it's a GNU/Linux router, you have options. MRTG is the obvious
choice, but you can always brew your own with RRDtool or GNUplot/sar.

Mark