Bandwidth vs. Latency on a T1 line

T

tdowning

Gawd
Joined
Oct 7, 2000
Messages
518
Please forgive the wall of text, but I think I should start at the beginning:

I'm volunteering for a non-profit that has a T1 WAN backhaul.

Normally, I can ping "www.google.com" and get ping times Minimum=15ms, Maximum=150ms average=18-50ms, depending on how many other users are using the web.

However if downstream bandwidth (From remote location to us) becomes saturated, ping times spike to 400-800ms.

I'll be the first to admit, I have loads of "Book-learning" so to speak on networking and computer technology, but first-hand, practical experience, particularly with WAN technologies, not so much. I have been subscribed to the "Security Now" podcast from Leo Laporte's Twit.tv podcast network, and remember Steve Gibson devoting a whole episode to the "Buffer Bloat" phenomenon.

If you are NOT familiar with this issue I'll refer you to the wikipedia page at http://en.wikipedia.org/wiki/Bufferbloat

TL:DR summary of buffer bloat:
More (Larger capacity) buffering by a router of packets that must go ouy over a low bandwidth network link is a bad thing. It causes packets to wait a long time (relatively speaking) to go through, and this can cause issues with timing sensitive communications, (Online gaming, like WoW, VoIP, Web Browsing,) and other real-time communications.

IIRC, Steve explained that excessively buffering packets causes TCP flow-control to break down, which can further exacerbate the problem.

With all of that being said, now, what I am asking of you is this:

1. Am I completely off base here? The discussion in the Security Now podcast was referring to Home routers where saturating the more limited upstream bandwidth caused reduction of available downstream bandwidth.

2. Am I correct in that saturating downstream bandwidth causing these issues means that the issue is on the provider's side of the T1 line, because that is where the bottleneck on the downstream side is.

3. does anyone have a T1 or bonded T1 that they can test like I have tested ours, (saturating upstream or downstream bandwidth and comparing ping times to a reliable major website.)

4a. Does a CSU/DSU count as a "Hop" across a router, and would it normally have an IP address (or two since it's a router)?

4b. If I traceroute www.google.com over our T1,
  • Hop 1=192.168.1.1=Our Nat Router
  • Hop 2=an IP adjacent to our public IP conforming to a /30 network. Ping is 1ms regardless of T1 utilization
  • Hop 3=the source of the trouble. 9ms ping when T1 is idle, 400+ms when downstream is saturated
If I am interpreting this correctly,
Hop 1's IP is the LAN interface of our NAT router, a 100Mbit Ethernet link from my computer to the NAT
Hop 2's IP is the 10/100baseTX interface of the CSU/DSU mounted on the wall near the NAT router.
Hop 3's IP is the T1 side of the CSU/DSU on the providers end of the T1 line, and this is where the choke-pont is, where the remote CSU/DSU has a 10/100 or greater link in, but only a T1 going out.

Am I interpreting the above correctly, and is the above typical, and/or standard operating procedure for a T1 internet link?

5. Am I correct that if properly managed the ping time to google.com should be more like 50-100 MS with occasional timeouts (maybe 10-25% loss, from replies being dropped by the remote CSU/DSU

6. From the traceroute, I can ping hop 3 from my phone, (via Verizon 4G and get consistent, reasonable ping times, 50-60ms regardless of T1 utilization. Pinging hop 2 from my phone will be consistent with what I'm getting at that moment, pinging google.com from the T1 line. Am I correct in that this is further evidence that the issue is on the provider's end of the T1 line?

Thanks fort your help,

Tim D.
 
1. no not really... saturating any bandwidth, up or down, will cause higher latency... the point was made that it can be easy to saturate upload because on your typical DSL and cable connections with asymetrical up/downs, they don't give you much upload at all... certainly not in the modern world of P2P

2. T1s are QUITE slow... your average cable speeds these days probably triple a T1s bandwidth.... 5mbps cable download is considered slow nowadays, now imagine if it was 1.5

3 doesn't matter what kind of connection it is

4a that hop is your ISPs gateway, sometimes a CPE, in your case with T1, most likely some equipment in your central office, the short answer is "it depends" on your delivery technology and ISP configuration

4b that's just because you're not fully utilizing the bandwidth of that leg of your connection... the shortage of bandwidth (your T1) is not that part of your route

5. there should be 0% loss on a reliable connection, that's the very definition of reliability, when you say "properly managed" you're starting to talk about either 1. not using your full bandwidth, or 2. using Quality of Service to prioritize traffic... and if your ICMP (pings) is what's prioritized, then yes, it should always remain low latency

6. you've saturated your 1.5mbps T1 connection, it's not hard to do (these days youtube on a single computer can easily do it) and is not surprising... it really is that simple... either secure a faster connection or use QoS to make things look better to you by masking the real problem

or hey... just tell us why you care that your ping times are so high, unless you want to do something to arbitrarily make the results of your troubleshooting look better
 
That said, every protocol does have a timeout so if you have lets say 50 clients trying to upload data you probably want to prioritize different kinds of data because some packets will be dropped due to the amount of total bandwidth. Depending on the amount of clients you might want to look into getting a small web proxy that caches data and possibly including Windows Update.
//Danne
 
Having investigated the possibilities, it looks like we will be replacing the T1 service with an AT&T business Uverse service, at 15/1.5 Mbit.
 
tdowning said:
Having investigated the possibilities, it looks like we will be replacing the T1 service with an AT&T business Uverse service, at 15/1.5 Mbit.
Click to expand...

Look into at&t fttp service for businesses. It should be a comparable price to your t1 and be many times faster.

Don't get me wrong, I like the uverse product but for a business it either works great or it doesn't.

If you need the upload, go with the fttp.
 
QwertyJuan said:
Yeah T1 was a really FAST connection... back in 1999. :)
Click to expand...

Yeah, I was contracted to install some routers for a company that was moving to T1. Had two techs from the company, one from Level3, and one from the router manufacturer on the phone all testing the connection after I racked the router. Took them 30 minutes to realize that the reason they were getting bad latency on the line, was because all three of them were running network testing utilities at the same time from 3 different datacenters with 10+Gbit connections and it was maxing out the T1's bandwidth.
 
rflol.. yeah... though arent ping times for a t1 still pretty amazing? i guess if you did nothing but gaming, whats the best to have?
 
Zohar78 said:
rflol.. yeah... though arent ping times for a t1 still pretty amazing? i guess if you did nothing but gaming, whats the best to have?
Click to expand...

Where I am a typical residential connection has a ping of 50ms to their ISP's "local" router, which means I almost never see better than 60ms to a game server. That T1 line they were testing was getting 14 - 20ms cross-country, over Fairpoint's crusty old copper (some of the wires still have tar and cotton insulation FFS).

I guess it might be worth it for gaming if latency really bothers you, but there are more than a few posts in the speed test thread showing residential connections under 15ms.
 
You must log in or register to reply here.
Back
Top