dslreports logo
 
    All Forums Hot Topics Gallery
spc

spacer

Search Topic:
uniqs
1123
share rss forum feed

TheHox

join:2012-05-31

1 edit

Upstream requirements for apartment building

Many of you have seen my posts before about an apartment building I've done some work for.

I am not their ISP, but I did install a PTP link to serve the building as it was in an area where the cable company did not have cable ran to it yet. I beamed over a 100mb/5mb connection with some nanostation loco m5s. I get anywhere from 80mb-100mb downstream and usually max out the upstream at 5mbps when doing tests.

Well there has been complaints about the internet service being poor and unreliable, slow and pages timing out. (Only during the school year when its fully rented. I talked to a student there today and he said it has been great all summer long) The owners apparently went and waved enough money to the cable co to have them start running lines to the building. I saw them today doing underground boring running conduit. I talked to the owner and he said he is having them install the 100mb/5mb connection in the building, thinking that is going to fix all the problems.

What is your opinion on a safe upstream to serve an 80 bed college apartment building?

I'm trying to figure out where my bottle neck is..

Upstream?
My routers? (Mikrotik at both ends)
The docsis modem?
My queues?
Too much porn?

jcremin

join:2009-12-22
Siren, WI
kudos:3
Do you have any idea what the general usage of the line is? With 80 college students, they could easily max out the 100 meg download, but I suspect the 5 meg upload will become an issue far sooner without some traffic shaping to keep a single person uploading a torrent from bringing the whole connection to a halt.

My WISP network (without too many power users) doesn't even hit 50 megs download, but I'm reaching 5 megs upload, and that is with each account capped with a much slower upload speed.

TheHox

join:2012-05-31
My current queues are setup to limit each 4 bedroom apartment to 10mb/2mb.

Looking at my MT graphs, it doesn't show a peak of over like 11mb, but that is not accurate. I've done multiple speed tests and large downloads there of at least 80mbps.

But even during the busiest times I've watched the router live real time and not seen it go over 50mbps downstream.


John Galt
Forward, March
Premium
join:2004-09-30
Happy Camp
kudos:8
reply to TheHox
The poor page load performance may be related to the users computer. If they do not flush their cache and history, defrag, etc. on a regular basis, then their system will be slow.

Do you have any type of connection limiting per user in your router?

TheHox

join:2012-05-31

1 recommendation

I have a few tenants that are going for degrees in computer science that know their stuff. They/I have both confirmed the slow performance not just their PC. But I won't disagree that some of the complaints we get can be from many other issues like that.

I just had un-throttled this kid and had him do a speed test.


I have each apartment on its own VLAN interface, then I limited each vlan interface with a simple queue of 10mb/2mb.

The only other rules I have is some custom rules that blocks torrents.

jcremin

join:2009-12-22
Siren, WI
kudos:3
reply to TheHox
What are you seeing for upload usage on your graphs?

TheHox

join:2012-05-31
Click for full size
Unfortunately I think these graphs suck.

I just had a kid to that 50mbps speed test, and it doesn't show anything of the sort, because they are only factored on a "5 minute average".

Is there a better way to monitor real time stats?

raytaylor

join:2009-07-28
kudos:1
reply to TheHox
If i remember right, to acknowledge 3.5mbps of incoming packets, you require 128kbps of outgoing packets.

So i would assume your upstream could be

I could be wrong here on my thinking. In NZ with ADSL, the ISPs where I live offer a full speed download and 128kbps upload, vs a full down and full up. The FS/128k plan is only capable of about 4mbit down because the 128k up limits the acknowledgement packets.

So by my thinking, if you are downloading at 50mbps, you will be using 1.6 megabits of your upstream.

Add in also the fact that your upstream is probably congested, it will mean that requests for webpages cannot get through which will be limiting the ability for a download to even begin.

My suggestion would be to limit each apartment to a 320kbps upload limit, or use the microtiks to equally balance each apartment so they have an equal share of the upload speed whenever it goes above 3mbps.


twizlar
I dont think so.
Premium
join:2003-12-24
Brantford, ON
kudos:3
reply to TheHox
You need to bump up the upload speed. I wouldn't supply less than 20mbit with 100mbit downlink. Looking at the ratios you have applied to each user, its way out of whack.

On your graphs, you should change the graphing frequency to 5 minute samples for the monthly. This will give you a much clearer picture of what your network is actually doing.

wirelessdog

join:2008-07-15
Queen Anne, MD
kudos:1
reply to TheHox
I would imagine layer 7 shaping would fix the issue but you would need some real horsepower to make that work.

mj3431

join:2003-04-21
STL, MO
reply to TheHox
Look to Cacti for more accurate graphing. I used MRTG for years and Cacti paints a much more accurate picture. Graphs update every minute with weather mapping capability we cycle on a TV in the office. I see every backbone link every 3 minutes.


TomS_
Git-r-done
Premium,MVM
join:2002-07-19
London, UK
kudos:5
reply to TheHox
Ive used the free version of PRTG to do more so real time graphs in the past.

Of course, you could also whip up a script that uses RRD tool to do the same thing for free.

PRTG is also a Windows based tool which may or may not be useful (which is also the reason why I wouldnt use it more extensively than the occasional exercise.)

But you are right. If you download 50mbit/sec for only a couple of seconds, you probably wont see it on your 5 minute averaged graph.

You need to download full speed for 10 minutes, or enough to cover 2 polling periods to get an accurate figure.

kf6ytc

join:2002-03-26
Turlock, CA
reply to TheHox
most cable modems crap out when too many connections are open, I first used cable for my wisp and once I reached about 100 customers the modem would just crap out and freeze but the total bandwidth wasn't being used, it was due to all of the connections coming through... one person could max out the modems/cable capacity all day long but if 100 of them used 25% of it, it froze up and caused slow pages, loading, etc

I would bypass it and do a torture test myself and always worked great... Finally we did a speed test using Iperf with several hundred connections and froze it up real quick...

It isn't a SMC modem is it?

TheHox

join:2012-05-31
Yea its an SMC they issue