Slashdot Mirror
How to Test Your T1?
lawpoop asks: "We have a T1 line for our building with a local ISP. Right now, we're looking for competitive bids from different companies. The local guy is offering a good price, but the larger guys are saying he may be overselling the T1 service through a DS-3. He swears he's not. So, how do I tell? The sales guys say 'There's bandwidth meters on the web,' but they fail to mention exactly how I can tell if I have a true T1. I've tried a half-dozen bandwith meters on various websites, and the results are highly variable. We've gotten 300-900 Kbps. Each site has disclaimers as to internet traffic, time of day, etc. Furthermore, we split the T1 out over a hub with two other tenants in the building. I'm coming through from behind that hub. How can I tell for certain that I'm getting a full T1? A service tech with a line tester? Any dead-on bandwith meters? What would an oversold T1 read out to be as compared to a true T1? If the larger guys are trying to scare me to their service with stories of oversold T1s, I need to know that they aren't doing it also!"
MRTG
24 channels of 64 kbits apiece. We sell T1s to customers, and if one of them wanted a util to test their bandwidth (the full 1.544 mbit) they could download a file from an ftp right at our pop. Or, have them ping flood you... use something like mrtg to graph the results, etc.
FLR
If you're getting T1 service, it should be possible to borrow a CSU/DSU, put it in remote-loopback, and make sure you have a full T1 of bandwidth. Or, if you own the router, you can just look at the statistics...
If you're talking about getting ISP service with "T1" equivalent bandwidth, that's a different story. You wouldn't be able to tell if the guy has "oversubscribed" you unless you find other buyers of the sevice and generate enough traffic to load down the DS-3.
To prevent getting burned, make sure your SLA clearly states the bandwidth you are expecting, and the means by which that is measured.
Cytlid has a good point - you get a T-1 from the phone company (or a reseller/CLEC) and it either IS a T-1 or it IS NOT.
I suspect that you're asking how you can tell whether or not your ISP is selling 50 million T-1 lines when he himself only has a T-3 connection with the rest of the world.
I think the simplest way is to ask. Talk to the sales engineers who work for the larger guys - tell them "Ok, you're trying to scare me away from a smaller vendor...how can I prove for myself how he's configured?" Ask the small guy "This looks like a really good deal...can you demonstrate to me I'll get X level of performance?"
Well, usually, you're buying a T1 from your location to the ISP or hosting company or whoever. 99 times out of 100, you're going to have a full T1 from your place to theirs (ie: 24 channels of 64kbps, or 1.544kbps). The 1% is most likely going to be some unscrupulous ISP; I've heard stories, but never seen a T1 sold to a customer that wasn't a full on "T1".
Now, as for your bandwidth, that's a different story. It is accepted practice to have oversubscription on your network; ISPs simply don't have the money to provide a full, balls-to-the-wall, 1.544mbps connection to the net for every single one of their customers. Local loop charges for simply a T1 from their office to yours starts at around $200 (in Alabama), and that's only if you use a CLEC. The bandwidth is what you're wondering about, and quite frankly, without having someone in the know inside your ISP, you will *never* find out how oversubscribed they are. What you pay above and beyond local loop charges are bandwidth access charges; you're actually paying for internet access at that point. Roughly analogous to paying for a phone line (local loop), then paying for dialup internet service (the T1, in this analogy).
Basically, unless you're buying a DS3 (44.762mbps) or above, you will never ever get committed, 100% full bandwidth on demand all the time.
I think you found the way, but you forgot to include the address for your site.
Slackware: old school feel, new school gear.
Its just a matter of how much it is over sold. If the "Big Guys" are try to tell you they dont over sell they are lying. Everyone does, playing the averages are how this bussiness works. Ask to see some MRTG graphs of thier gateways b/w. See if thier heads are bonking against the top very often. BTW chances are that the small isp will treat you very well while the super available mega corp sales man soon be replaced with a touch tone menu. Ask for refrences and call them.
The real test, frankly, is to get bandwidth from someone with heavy-duty backbone connection (e.g. AT&T) and just plain hammer it with mondo file transfers scattered across the day. If your transfer times are varying with Net traffic periods, your ISP is the bottleneck.
You might be able to get similar information cheaper by doing repeated traceroutes and logging the delay between the ISP and the next router up, which would indicate the queueing delay at the ISP's routers.
Lacking <sarcasm> tags,
An ISP that I know fairly well (*cough* work for *cough*) oversells bandwidth. They use mrtg (as has been suggested elsewhere) and any time a network segment reaches 80% utilization at any point in the day, three days in a row, that segment is upgraded.
Seems to work quite well to me, but maybe I'm biased. Try an get a conference with the techs (see if you can talk to their network monitoring team) and see if they employ a similar practice.
It's an old ISP myth. Everybody "oversells" their connections at some point in the stream. In the early days, this sort of thing was an issue, a small ISP would but a certain upstream bandwidth (usually one or two connections) and then sell pieces of it until they had sold more downstream than they had upstream. In practice, it worked well, since few people ever use their maximum bandwidth constantly. A few were fastidious about buying upstream bandwidth in exact proportion to what they sold downstream. They mostly went out of business or were bought by Verio.
Today, few ISP's actually have upstream bandwidth equal to what they've sold downstream. And it gets even more complicated when you consider that there are usually multiple routes out of an ISP, some of which can be easily overloaded, others less so, depending on where the traffic is destined.
The only worthwhile measure is a subjective one. Can you get 1.5Mb throughput on ANY site? On some sites but not others? Do you think you're going to get better service from somebody else? There's no exact answer as to whether you're getting your money's worth; experienced net admins have used a several connections over time, and usually know within a day or two whether they're on a good one or not.
First of all, if you have *the whole T1* then you should have control of the CSU/DSU and the router at your end. Accept no substitutes. If the guy is, in fact, splitting a T1's worth of bandwidth off a DS-3 (say) then you should (at the least) have monitoring privileges on the router and DSU, either via web, SNMP, or telnet.
As mentioned earlier, the DSU should show that your connection to the line is using 24 x 64Kbps channels, for a total of 1.544Mb/sec (minus a few k for channel overhead gives you 1536). ALL of those channels should be coming out your end of the CSU/DSU.
If you have control of the router, then you might try (again, as mentioned earlier) ping-flooding the router at the other end of the T1 and checking how many packets/sec get through, then multiplying by your packetsize. That should give you a reasonably close answer.
If the guy isn't bullshitting you, then he should have no problem giving you read access to the SNMP MIB on the router and letting you watch traffic - if you can access this, you can run bandwidth monitors like MRTG.
Good luck.
A hero is someone who knows when to run away. I am a hero. -Trent the Uncatchable
3 things:
1. tell your boss to fire you, and hire someone with clue.
2. you have a router, use it, dipshit.
3. see #1 and #2.
Jesus christ, do you think the router's there just to look at?
S.L.A. (Service Level Agreement)
If said small provider is telling the truth, then he won't have a problem signing one. I've found in my area that the big guys are the bullshiters when it comes to SLA's.
"Klaatu, verada, necktie!" -Ash
This even applies to phone lines.. That's why It's sometimes hard to get a call through on Mothers' day. The phone companies provision to handle 99%+ of the volume spikes, and mothers' day can consistently make it into that last percentile. Even so... handling 99% of the traffic spikes still comes to far less than one circuit for every two subscribers.
To test if your T1 is really a T1, you can try setting up for an FTP (or whatever) session with your buddies... Change things to make sure that you've each got the route to the other going through the first hop on your T1. As long as that router doesn't icmp-redirect the traffic, (in which case your measured bandwidth will be more in line with local ethernet traffic) you should be able to get a good estimate as to the raw bandwidth of your 'last mile'. (it'll actually be the lesser of your uplink speed and your downlink speed minus a little bit)
The next hop would be to set up a transfer with something inside of your ISP's primary network. (did they give you a web site on one of their systems, etc??). That'll allow you to test for local bandwidh bottlenecks and give you a theoretical maximum to the outside world.
The last link check would be to find a machine on a fast network that's not on your local ISP's but is (topologically) close. Try doing traceroutes to nearby universities.. See if you can find one that doesn't put you through 3 different ISPs. Then try and transfer data from/to them and see how fast it goes.
You'd be best to try connections to a few semi-local sites. Otherwise it'll be hard to tell if a low bandwidth reading is the fault of your ISP, or the server's ISP.
It's pretty much useless to check bandwidth to random (distant) sites. Once you get a site that's a reasonable ways away (topologically or geographically), then you run into the vagraties of internet traffic (see the article earlier this week about 'net quakes)..
BTW: When I speak of being topologically close, this is different from geographically close. I remember one case where getting a packet to a machine not more than 100 feet away (but on another ISP) sent traffic from Vancouver, down the coast to Silicon Valley and back. Needless to say, ping times stank. In that case we were geographically close, but topographically distant.
THere would be two times to test these transfer speeds: Low time (e.g. 4am) and prime time (Last time I peeled apart ISP traffic stats, traffic peaks were around 8-9pm for home traffic and about 4PM and 9AM for commercial traffic)
Free Software: Like love, it grows best when given away.
I thought I had a T1 but it kept asking if I had seen John Conner, turns out I got a T1000! Word to the wise NEVER shop with Cyberdine!
-Jason
That's it. I'm no longer part of Team Sanity.
T1s don't use hubs, or switches. They aren't ethernet.
As was mentioned in an earlier post, a T1 uses something called a DSU/CSU to manage the 24 BLines on your T1. This is sometimes built into a router, on a card in a router, or an external device running to some sort of high speed serial line (not always HSSI though) that can be linked to your router.
You can check the settings on this DSU/CSU to make sure you have a full T1 and not a fractional (all 24 channels is a full T1, less than 24 is fractional), but that won't help with finding oversubscription at the other end. There's really no easy way to check that, but if you never notice, who cares? Just make sure you get an SLA for the bandwidth you expect (usually the full 1.544Mb/sec on a T1) and if you at any time are unable to get that due to oversubscription by your ISP (all of them do it, some more than others though), you are entitled to compensation (often a partial refund or even being paid).
Couple of things, first, all ISPs oversell bandwidth. I am assuming the 'larger' guys meant that the local ISP likely only has a single DS-3 to his upstream provider. Depending on how small he is, this may not be an issue (/all/ providers oversell at some ratio, it just depends on what that ratio is that determines if the provider sucks or not) depending on how many other customers he has, what types of circuits your providers customers' have, and what type of customer they are (business vs. residential), etc. Obviously there are other factors too.. if your provider suddenly grows and doesn't increase its upstream capacity, thats an issue.
/most/ circuit IDs from /most/ telcos is something like XX.AAAA where AAAA is frequently what your concerned about to determine the circuit type, but the format of the CID depends on your telco (there are many guides out on the internet at decyphering these to determine what type of service they are, or you could call your telco and they could tell you.. maybe.. if your good.. heh)). The CID should be on the smartjack..
/can/ be, without having to deal with other provider's networks being congested, etc (the traffic in this case will be local to your provider's network, so if thats congested then um.. that sucks).. You should get somewhere around ~192KB/s.. if its slightly less don't worry about it, there is some overhead involved, etc.. When you do this be sure you only have /one/ machine connected to test (or you can verify there is nothing else that is generating traffic that is going over that circuit, etc... don't assume.. check. (there are many, many, many tools to do this..) to see whats hitting the ethernet interface of your router (its a lot easier to check if you have access to your router, as you can just do a show int on a cisco to get traffic statistics, etc).
Now, on to determine if you have a 'real' T1. Many providers tend to sell frame relay service as a 'T1'. While its true that the circuit itself from the telco to your place of business is a T1 (unless you have say 56k DDS service, etc), after your traffic hits the telco's switch, it transverses their frame relay network, and eventually gets sent to the frame host on your provider's network. This can suck for many reasons, however the biggest one is that the provider can get away with purchasing a CIR that is less than 1.5 mbps (like 768kbps), and just have it be able to burst up to 1.5. This can suck a whole lot if the telco's frame network is congested, and you can never burst, and/or you constantly want to use more than the CIR etc..
You can tell what kind of circuit it is by quite a few ways, if its installed and you have access to the router you can simply check the encapsulation on the T1 (if its set to frame relay, its a frame circuit, if its set to PPP or HDLC its a point to point circuit), you can also tell by the format of the circuit id (the first part of
If you have a frame circuit they should be charging you quite a bit less than a point to point T1, especially if the CIR is low (the lower the CIR the less expensive generally). Point to point T1s are preferible in almost all cases unless your worried about cost.
So now that you've determined the type of circuit you have, you can check what speeds your getting. Its usually best to do as other people have mentioned and download (and UPLOAD too.. you should check both speeds) from an FTP server on the provider's network. This will give you the most accurate picture of how fast your connection
You should also do a separate test to a major site.. You could download the 1.4.0 Java SDK from Sun for instance (that should give you a decent speed). Don't worry if this is somewhat lower than 192KB/s, as that can be caused by congestion in a network inbetween you and sun that has nothing to do with your provider, etc.. If its consideribly lower than that speed consistantly (and other sites with bandwidth to spare yield the same results), then I would contact your provider about it. It could be that your provider doesnt have enough upstream capacity, or about a billion other things, but they might be able to tell you any known issues, or that the problem isn't them and/or tell you the current utilization levels of their upstream circuits.. heh
Those who test by FTPing large files and watching the transfer rate, should understand these limitations (kindly explained to me by J.Spencer Love).
I had a similar problem trying to host a large-bandwidth video clip. It turned out the bandwidth of my 10Mbps line did not saturate at all (in fact, it was utilized at mere 5%), so neither did the trans-Atlantic connection. The bottleneck was the internal buffer in client and server software.
This also means you may not need that much bandwidth to push the speed of your FTP/TCP-based tasks to its limit.
17779 eligible voters in a district, 17779 'vote' as one. This is Russia.
Write up a short story that says your company uses Linux because they hate Microsoft. Then, place that story as an HTML page on a a webserver on that T1. Submit the story to Slashdot. (don't worry, it'll get posted.)
If you're not getting enough hits to it from Slashot, put a small Flash animation on the page. This will infuriate the users, they'll bitch about it and cause more people to hit your site.
If your server catches fire, then your T1 is up to speed.
What the competitors are telling you is that this particular provider has a DS3 circuit that has basically 45 Mb/s of throughput. In other words it can support 29.22 T1 users running at full throttle. In reality if you are using your full bandwidth then you are not!!! VERY IMPORTANT HERE!!! If your T1 is loaded then the routers on each end are dropping packets since they don't have very big buffers. This causes packets to be retransmitted, messages to be sent requesting TCP connections throttle down, timeouts, etc. Remeber that the real goal is how many WWW pages, email messages, FTP files, etc. that you can move. It is NOT how many bits per second you can send! If you are running at about 90% capacity then you can consider the line MAXED out. The same thing goes for your provider on his DS3 line.
If you throw in the burstyness of TCP/IP traffic your traffic really maxes out bouncing around somewhere between 65% and 90% of its max rated load. When you add in the fact that people like to have headroom and the size steps between T1 and T3 or partial T3 nobody uses their max bandwidth all of the time. This is something that your provider uses to his advantage by selling more than 29.22 T1's. In reality that provider may have 1000 dial up customers, many more DSL customers, people with dedicated 56K connections, and maybe even some old ISDN connections on top of the other T1's that he has sold. It is fairly safe for a provider to oversell a connection somewhere between 5x and 20x. Especially if you consider that people that have DSL's usually leave them on but don't use them for that many hours in the day. Headroom is defined as being prepared for the slashdot effect!
You need to figure out if your provider speciallizes in retail (home) customers or business customers. If it is home customers then his load will spike in the evening when poeple get home from work and on the weekend. If it is business customers then his load will spike during business hours. Also consider if your provider is hosting very much traffic beyond home users personal WWW pages. If he is then that is bandwidth that is not available to resell.
The only real way to tell is to look at your provider's router logs. Don't just look at averages for a day! Look at averages generated at least every hour over the course of at least a week so you can see when his network (which you will be a part of) loads up. If he consistantly stays below 80% of his upstream bandwidth and will upgrade his upstream connection if it passes that then you are fairly safe. On the other hand if he routinely pushes his max then that is exactly what his competitors are warning you about! If he loads up then everyone downstream from him will slow down as that will become the bottleneck.
Another thing to consider is that it sounds like he has a single DS3 upstream connection. Ask him if that is true and if he has any plans to become multi-homed. This has two major advantages over a single fat pipe: redundancy and load balancing. If he connects to two or more of the backbone providers then the traffic can most likely pick the shortest route to its destination while still having the other one available if one of the DS3 lines goes down. I'm sure the people @ /. have more than one connection if they are served from their business. It is more likely that they are hosted by a large hosting company that is already multi-homed though. A final thing to ask is if this provider does any traffic shaping. This can throttle users that are consuming more than their fair share. But if you are paying for a T1 you should get all of it. Sometimes though a customer will have a partial T1 (this requires a full T1 connection from their site to the provider though) and get to use any excess if it is available.
This is what I used to do for a living so I know a little bit about what I'm talking about! Once again his logs can tell you if he has oversold his service as his competitors suggest. But put some value on a provider that has more than one connection to the upstream Internet regardless of how large their single pipe is. If you want to have some fun ask him how much an OC196 would cost. 8-)
Regards, Tres
Restore America: Dr. Ron Paul for President!
... why do you really care? If you're getting the service you need, why bother? If you're satisfied with the current bandwidth and ping times through your T1, why buy something more expensive even if it is faster/less congested/has blinkity blinky florbs?
All reasonably priced providers will sell you shared capacity. And overselling is (usually) a question of what the other customers are doing and at what point your upwards link is going to get upgraded. You'll never get full bandwidth connectivity to the providers peering point for yourself unless you're willing to pay through the nose.
But do you really need that or are you satisfied? If you're satisfied, spending more is a waste of money, wether or not your current provider is using IP over avian carrier to connect you.