FTP running slower then it should be....

[Runes911]

I have a colocated box that I set up an ftp site on. At first I used serv-u ftp and all the downloads went at 80k a sec from the server (my servers connection is 2Mbps up/down and it bursts at 10Mbps and my cable is 1.5 Mbps down) so I checked to make sure that there was no download limit, and there wasnt. So I uninstalled serv-u and installed BulletProof FTP to see if serv-u was the problem, after it finished I tried a download from it and it is STILL going 80-90 k a sec MAX is there something I am missing, should I be useing a different program to run the FTP server? The only other program running on the server is HL dedicated server but it was empty. The machine is running win2000 w/SP3.

 

[JackMDS]

Most Internet cable connections are slower or have restricted Upload (usually 128Kb/sec.) When someone takes files from your server it is using the Upload feed of your connection. 80KB/sec. (B=Byte) is usually an Upload of about 800Kb/sec. (b=bit).

 

[Runes911]

This is a colocated server that is running off of a 2Mbps symmetrical line (2Mbps UP/ 2Mbps down) that bursts at 10 Mbps I am not running this off of my cable connection.

 

[spidey07]

not too far off realy. 80 KB (KBytes)/sec would be 640 Kbs, then round up to 750 Kbs for overhead/latency. Depending on your provider however the "burst to 10 Mb" could be completely bogus.

You're downloading off a cable connection meaning your upload could be capped. this also affects your ability to download. what OS are you using to check your ftp speed?

 

[Runes911]

I am useing windows xp and I have had serveral others test my FTP. OK I am gonna explain this in more deatil.

The box I have colocated (FTP server) has 10 Mbps (10Mbps= one MEGABYTE per second when converted) which is SHARED among 4 other computers THEREFORE I get 2Mbps or 250KB per second when ALL of them are using the connection, when all are not connected I get the full 10 Mbps or 1 megabyte per secound.

My CABLE LINE is 1.5 Mbps/128kbps WHICH MEANS I SHOULD BE PULLING 150 KB/s which is my MAX DOWNLOAD from my FTP which has a MINIMUM of 2Mbps UPLOAD.... 2Mbps=250 KB/s

I am trying to make this as clear as I can it keeps seeming like no one understands please let me know if I need to explain more.

To sum it up:

My Server: 250KB/s
Client 150KB/s

Client is NOT recieveing full 150 KB/s and I cannot figure out why.

 

[spidey07]

SHARED among 4 other computers THEREFORE I get 2Mbps or 250KB per second when ALL of them are using the connection, when all are not connected I get the full 10 Mbps or 1 megabyte per secound

I'm kind of scratching my head here, that's not how a network works. If there are other machines using the same bandwidth they can very well completely overwhelm your service leaving you with poor FTP performance. All four machines are competing with each other, not playing nicely and using exactly 1/4 of the 10 Mbs. What are the exact details of your colocation service specifically bandwidth related?

I'm trying to look at this from both sides of the connection.

Problems on server side
1) speed/duplex mismatch on NIC/switch. Check the NIC and make sure there are no errors like alignment errors and CRCs. Ask your provider what you NIC should be set to - auto or force speed/duplex
2) bandwidth. maybe ask your provider if this is normal. Also the other machines could be monopolizing your bandwidth.

Problems on client side
1) cable upload speed, this can affect your download speed
2) latency, when you ping your FTP server what are the round trip times?
3) mucking with registry settings. Defaults for XP IP stack are very good and don't need to be changed.

hope this helps.

 

[Runes911]

Taken directly form my COLO providers site:


8) Can you tell me some about your network and how you limit bandwidth?
Our Internet access consists of an Ethernet to fiber connection. Our bandwidth is delivered via a single RJ45 jack into our datacenter, plugged into a D-Link DES 3326 Layer 3 Switch.

Colocated computers are either:
Uplinked directly to the DES 3326 (one computer on a 10 Mb/s full duplex pipe),
Uplinked into unmanaged bandwidth distribution switches (shared 10 Mb/s full duplex pipes), which in turn are uplinked directly into the DES 3326, or
Uplinked into customer provided distribution switches (dedicated 10 Mb/s full duplex pipes), which in turn are uplinked directly into the DES 3326.
A minimum of Cat5 rated cabling is used throughout. No hubs are used to distribute The Net Gamer's connection.



10) How does the shared port work for colos?
It's actually quite simple. On our main distribution switch, we have several dedicated ports manually set to 10 BaseT full duplex transmission. Connected via patch cables to those ports are unmanaged bandwidth distribution switches. It is into those switches that shared colo computers are uplinked. That's why your NIC shows 100 BaseTX connectivity, but rest assured, your equipment only has full duplex access to 10 Mb/s of The Net Gamer's bandwidth.


12) How will this load balancing work in practice?
Let's assume for a moment that there are five computers all hooked up to a 10 Mb/s full duplex port. Four of the five are for all intents and purposes idle, the fifth is serving up files at the full 10 Mb/s ceiling. From your 2 Mb/s cable connection at home, you begin a download off of your computer colocated with The Net Gamer. Over time, your download will get to the full 2 Mb/s speed your line is capable of, bleeding the 2 Mb/s from the computer that was serving files at 10 Mb/s. If only your two computers are active, you and he could theoretically serve up files at an aggregate of 5 Mb/s each, no problem. As more of the colocated computers on your switch begin to access the bandwidth available to all of you, your transfer speeds drop until all of you are hovering around the 2 Mb/s range. In all fairness, this load balancing comes at a cost. The cost is latency and packet loss as the shared port's 10 Mb/s ceiling is breached. When there are more requests than bandwidth available, the switch will react by using backpressure (latency increases) and buffer purging (dropped packets) to manage the flow through the port.

Also, I tested the line before and after on DSLReports and it reported I was getting 600KB/s down and 500KB/s up, yet I was only pulling 80KB/s. Maby someone knows of an ftp I could upload something to to test upload speed differently?

Also if you want to test it and see waht you get on it: 66.28.14.53 its an anony. ftp

 

[spidey07]

Unfortunately your colocation service doesn't understand network communications judging by those responses.

So either your 10 Mbs connection is saturated by your other computers or the colocations internet connection itself is saturated or possible upstream. I'm sure their connection isn't tier 1 so they could have upstream troubles as well.

I'd really be calling your colocation place to tell them you're not getting the performance you are paying for.

 

[Runes911]

This is what they have to say about their bandwidth utilization:

14) Ok then, how close are you to maxing out your line?
As of May 1, 2002, our peak transmit bandwidth utilization over the prior thirty days was 52.8 Mb/s.
As of May 1, 2002, our peak receive bandwidth utilization over the prior thirty days was 38.7 Mb/s.

This is a FAQ I am pulling this from it wasn't personal responces

I do know they are new to colocation but they rent out HL servers and they are pretty good so far. I still can figure out why I get a upload of 500 KB/s when I test to DSLReports yet my ftp is only getting 80 KB/s up... which leads me to believe that only FTP transfers are affected in some way....