Network throughput question

[steingra]

I am attaching a screenshot for conversations sake.

I have this free network testing tool http://www.raccoonworks.com/Products.htm that I have been using to test out connections between various speed/class of computers and the main sagetv server. The results are interesting for sure.

One thing that I have seen happen, I dont understand. If you look at screenshot, you will see that when I transfer the test file (its like 360,000 KB) from my laptop over a wireless connection, it is only using 16% of the available bandwidth. The wireless connection says its 54 Mbps, and an excellant connection. But why would it only use 16% of the bandwidth. That is what I dont get.

Now, I tested the same file download on several other computers. Both of them have 100 Mbps connections. So Windows reports anyway. One of the machines is on a long rj45 cable (100 feet?). And it reports speeds of about 50-60 Mbps. And it uses about 49% of the bandwidth (when looking at same task manager screen). It connects to the same 100 Mbps switch as the server. Also the CPU usage is not too high when downloading this file. Maybe 35% I think. So its not like CPU bound.

Now, the third test machine, is connected on to the same switch as the other machines (test machine 2, and the server) And it gets a 90-100 Mbps download speed. It flies!!! By far the fastest machine. And it shows about 90%-95% network bandwith being used.

I have not hard coded any network parameters, they are all at whatever windows set them to.

I would like at least see both hard wired clients to be at same speed. Not sure why its only half the speed. The network cards are different, so I am going to dig up another 3com card and stick it in there. And see if it makes any difference.

Just trying to eek most bandwidth out of the machines across the network both wired and wireless.

Edited: I attached screenshots of the network usage over the wireless connection in Task Manager. Still baffled why its only using 16% why isnt it using more? What settings can I change to force it to use more....maybe? hopefully?

And also attached a screenshot of the results of the speed test. So the wireless laptop is still getting about a 10Mbps connection.

[paulbeers]

I guess my question is, what is the problem? I understand you want maximum bandwidth, but there are sooo many variables for network connections. As far as HDTV the most you are going to transfer between a server and a client is approx 2.4 MB per second or 19.2 Mb/s for high def. You shouldn't expect much from wireless connections. Even tho you are connected at 54 Mb/s, you lose about half the bandwidth just in communication alone. Best you should ever really expect is probably about 25 Mb/s of actual file transfer from what I understand. Maybe someone else can shed some light on that. I know that I couldn't even get a sustained SDTV Mpeg tranfer on my "old" wireless network that was 802.11b, and that only transfers at 4.5 mb/s which running 11 mb/s should have been "no" problem if it had full use of the bandwidth. I guess this is always my question for someone with your "predicament", is it really a problem? Are you making a mountain out of a mole hill? Just because you aren't getting maximum transfer, does it matter? If you really want to maximize bandwidth, go buy all Gigabit routers and cards and then you will never have to think about transfer times. You probably are giong to in the end causing more problems and stress more about something than it really matters.

Just my two cents worth.

Oh and I have a 100 mb/s network and 54 mb/s wireless network. I will never run a problem like you mentioned before, because everything works fine and I will just end up tinkering and blowing something up. If it ain't broke.....

[steingra]

hello paulbeers

Thanks for the input. I am not trying to make a mountain out of a mole hill he he.... Im just trying to get all my client pcs to play my sagetv content as nicely as possible.

I dont want to give up on the wireless thing yet. I have seen that when using the VideoLan client, it only uses 4% of network bandwidth when playing a sagetv show across the wired network on this PC (~620 Mhz celeron). That seems to tell me if I used the videolan client, it *should* work on a 54 Mbps wireless network connection.

I have been sitting here watching the task manager for 30 minutes, and its staying at about 4-5% of network bandwidth when playing the mpeg2 file using videolan client. I AM VERY SURPIRSED that it is only using that much.

The video and audio sound and look great....

In general networking works great, you dont have to think 2 times about it, UNLESS you are trying to push large files, heavy database usage, OR video based applications. Just in my limited experience....

So this has taken on a whole new life of its own. Now I am looking at settings and options on the network cards, trying to figure out which ones might be worth changing (like forcing it to 100Mbps full duplex) but what is downside of it???

I know this is a huge question...but I think its a good goal. And I am learning alot more about networks/networking in general.

I guess the kicker is that when using the SageTV client on this pc, the video and audio quality is not as good as on video lan client which is free. Dont get me wrong, I like all the benefits of sagetv client, but I cannot put my finger on why the same hardware would act so much different.

SageTV client playing a movie = choppy video/audio mostly OK
VideoLan client playing same movie= looks and sounds great

[stanger89]

Originally Posted by steingra

I am attaching a screenshot for conversations sake.

I have this free network testing tool http://www.raccoonworks.com/Products.htm that I have been using to test out connections between various speed/class of computers and the main sagetv server. The results are interesting for sure.

One thing that I have seen happen, I dont understand. If you look at screenshot, you will see that when I transfer the test file (its like 360,000 KB) from my laptop over a wireless connection, it is only using 16% of the available bandwidth. The wireless connection says its 54 Mbps, and an excellant connection. But why would it only use 16% of the bandwidth. That is what I dont get.

That's because the connection "speed" is essentially meaningless, it's basically just what mode the connection is runing in and has almost effect on your maximum throughput. What you have shown, is exactly why most of us recommend not using wireless if at all possible, because you get nowhere near the bandwidth you think you would.

Now, I tested the same file download on several other computers. Both of them have 100 Mbps connections. So Windows reports anyway. One of the machines is on a long rj45 cable (100 feet?). And it reports speeds of about 50-60 Mbps. And it uses about 49% of the bandwidth (when looking at same task manager screen). It connects to the same 100 Mbps switch as the server. Also the CPU usage is not too high when downloading this file. Maybe 35% I think. So its not like CPU bound.

Well 35% seems high (maybe it's not I'm running about 30-50% transferring a file at about 250Mbps ) but there's really a number of things that can slow it down, your wire, your NICs, etc. Are they just cheap integrated NICs? Are they good integrated NICs? etc.

Now, the third test machine, is connected on to the same switch as the other machines (test machine 2, and the server) And it gets a 90-100 Mbps download speed. It flies!!! By far the fastest machine. And it shows about 90%-95% network bandwith being used.

That's actually really good, so what's different? Are you using different wire? Shorter lengths? Are these NICs different?

I have not hard coded any network parameters, they are all at whatever windows set them to.

I would like at least see both hard wired clients to be at same speed. Not sure why its only half the speed. The network cards are different, so I am going to dig up another 3com card and stick it in there. And see if it makes any difference.

Ah, that doesn't surprise me, I bet that would help. However if you're buying (not just "digging up" from your current stock). I would NOT buy a 100Mbps card, I'd go with something like this:
http://www.newegg.com/Product/Produc...82E16833106105

Just trying to eek most bandwidth out of the machines across the network both wired and wireless.

Edited: I attached screenshots of the network usage over the wireless connection in Task Manager. Still baffled why its only using 16% why isnt it using more? What settings can I change to force it to use more....maybe? hopefully?

There are lots of reasons, but probably connection issues, what I would try is moving the client/WAP closer together and eliminating any potential issues there. Could be your wireless NIC too. But just sort of speculating here.

[steingra]

I ended up using a similar 3com card, and the transfer rate is about the same as the other 3com card 90-95% utililization of stated bandwidth. So that was kinda wierd that just swapping out a different card would make such a difference. They were both 100 Mbps cards, but drastically different transfer rates, and thats without changing anything on the workstation. Just plug in new card and go run test.

I have read several articles where people seem to be getting higher throughput on their wireless connections using equipment similar to mine. 26 Mbps is the highest result I have seen so far. Which to me seems like it would be enough.

ALso, I ran the VideoLan client on the laptop using the wireless connection in the laptop (which is running at about 10Mbps for every test), and the video played OK. It was a 5 GB file recorded from sage (about a 2 hours show). I would get sick and tired of seeing my shows that way in a short amount of time. But it wasnt aweful...just not nearly as good as I was hoping for.

It did skip sometimes, but I was watching the task manager/network monitor and the % dropped down to like 0% for like 5 or 10 seconds and then would go back up to the 16%. I dont understand what is going on with the wireless...why it would chug along steady at 16% and then drop off to 0% only to go back upto 16% again for several minutes. SOMETHING must be causing it, just dont know what.

I feel like there should be some magic settings I can make on the router/wireless cards to make this work better. Maybe I am just asking for it.

Im not ready to give up yet on the hope of WIFI video streaming.

BTW, I hooked up an more sensitive antenna to the Dlink router, and it seems to have made overall speed and link quality better from what I can tell. I mean 10 Mbps of actual throughput inst bad for wireless, but I would like to get it twice that much. argh

[Karen0302]

steingra,

I run a rather large home network consisting of 7 PCs connected through 2 switches and a router. My systems run at speeds of 1 Gbps, 100 Mbps, 54 Mbps and of course the router to internet connection is at 10 Mbps.

It seems to me that there are several factors which affect network throughput. The speed of the NIC, the features on the NIC (especially at 1 Gbps), the quality of the cables used, and finally the speed of the systems themselves. Here are several examples from my own experience which may illustrate the points.

First, the quality of the cables. This happened several years ago. I needed to move one of the systems and the cable wasn't long enough so I ordered a new cable. When the system was back up I noticed that files were transfering noticably faster. I got curious and move the system back to using the old cable and transfer speeds dropped. Back to the new cable and they went back up. After this I never bought dirt cheap cables again; although I also did not go to the other extreme of buying really, really expensive cables like Oxygen free Monster Cables!!

Second, features on the NIC. When I first moved to a Gigabit LAN, I was getting less throughoput than I had expected. A bit of research taught me about a feature Jumbo Frames. A normal TCP frame is around 1500 bytes. A Jumbo frame can be up to 7000 bytes and when used can really speed up throughput on a network. After getting getting Jumbo Frames working on the Gigabit portion of my network, I transfer files at about 1 GB/minute.

Third, speed of the processor. Even with NICs that have features which offload creating and checking of the checksums, there is a lot of processor involvement in transfering files. Even on my Gigabit network, when I am transfering files to my main system (2.4GHz P4) the processor is usually running 95%+ during the transfer. If I transfer from one of the systems with a slower processor, it is usually running flat out and my throughput will drop accordingly.

As far as the Windows Bandwidth Usage report is concerned ... well, I never pay much attention to it. I tend to look at the real numbers. For that I use a program called NetStat Live from AnalogX. It is a freebie and gives the the actual network throughput in numbers. You may wish to look into this.

When I get some time, I will look at exactly what throughput I am getting on my laptop (it's been a really long time!) and let you know.

Karen

[steingra]

Originally Posted by Karen0302

steingra,
First, the quality of the cables. This happened several years ago. I needed to move one of the systems and the cable wasn't long enough so I ordered a new cable. When the system was back up I noticed that files were transfering noticably faster. I got curious and move the system back to using the old cable and transfer speeds dropped. Back to the new cable and they went back up. After this I never bought dirt cheap cables again

That is a good point. Currently, I made the long cable myself, to go all the way upstairs (maybe 100 feet or 150 feet at most?). That could be part of the problem (That I made it myself) . But even though I am not getting the full 100Mbps, I think the current speed on that cable should be acceptable. I just have to run more tests and use the sagetv client and videolan client, and see how it goes.

I will definatley keep that in mind. I might need to buy a long cable from someone, anyone you can recommned for quality affordable cables? There are sooooo many places to buy them online. That might get me the rest of the bandwidth back. And there shouldnt be any interference on the cable, its not really near anything to interfere.

Originally Posted by Karen0302

Second, features on the NIC. When I first moved to a Gigabit LAN, I was getting less throughoput than I had expected. A bit of research taught me about a feature Jumbo Frames. A normal TCP frame is around 1500 bytes. A Jumbo frame can be up to 7000 bytes and when used can really speed up throughput on a network. After getting getting Jumbo Frames working on the Gigabit portion of my network, I transfer files at about 1 GB/minute.

I heard of those Jumbo frames a while ago, but long since forgot until you mentioned it. Arent they only for 1 GB nic's? What is the process, and/or which software do you use for that? Dr. TCP?

If you can transfer files around 1 GB/min...now Im thinking I should just upgrade to 1 GB NIC's the cards arent that expensive. hmmmmm. What kind of 1 GB nic's are you using?

Originally Posted by Karen0302

Third, speed of the processor. Even with NICs that have features which offload creating and checking of the checksums, there is a lot of processor involvement in transfering files. Even on my Gigabit network, when I am transfering files to my main system (2.4GHz P4) the processor is usually running 95%+ during the transfer. If I transfer from one of the systems with a slower processor, it is usually running flat out and my throughput will drop accordingly.

Yes I have noticed that the CPU is definately more taxed, when I went from 10Mbps to the 50-60Mpbs file transfer speed. (CPU is ~620 Mhz celeron). I know its slow but its working pretty nicely with Videolan client to play my sagetv files.

So I definately need to consider that, if I ever planned on sticking a 1GB NIC in this class of a machine, I dont think the CPU can handle that level of file transfers. SO maybe just maxxing out the 100Mbps is best plan for this machine. One other thing I will do, is to physically move the machine right down next to the switch in the basement, and plug it in using an off the shelve cable, and that could eliminate other problems. The other machine I have plugged into the switch is working right at about 100 Mbps, but its using a 5 foot patch cable I bought somewhere.

Originally Posted by Karen0302

As far as the Windows Bandwidth Usage report is concerned ... well, I never pay much attention to it. I tend to look at the real numbers. For that I use a program called NetStat Live from AnalogX. It is a freebie and gives the the actual network throughput in numbers. You may wish to look into this.

I thought the Windows Bandwidth was fairly accurate, but its just not very full featured. I have been looking for various network testing programs. I did find the Racconworks program, which is very nice. But it doesnt show real time stats, just a nice after the fact bar chart. I will check that program out, thank you.

Originally Posted by Karen0302

When I get some time, I will look at exactly what throughput I am getting on my laptop (it's been a really long time!) and let you know. Karen

Thanks

[stanger89]

Originally Posted by steingra

That is a good point. Currently, I made the long cable myself, to go all the way upstairs (maybe 100 feet or 150 feet at most?). That could be part of the problem (That I made it myself) . But even though I am not getting the full 100Mbps, I think the current speed on that cable should be acceptable. I just have to run more tests and use the sagetv client and videolan client, and see how it goes.

I will definatley keep that in mind. I might need to buy a long cable from someone, anyone you can recommned for quality affordable cables? There are sooooo many places to buy them online. That might get me the rest of the bandwidth back. And there shouldnt be any interference on the cable, its not really near anything to interfere.

I'd be reluctant to think that the fact you made your cables is causing problems at 100Mbps. I run all "home-made" cables and I get about 250-300Mbps on my GB-E connection, and I think that's limited by the speed of my array.

You could try Ixia Qcheck to see what your throughput actually is. The problems with the methods you've been using (file copy) is that you could be limited by your HDD read/write spead (though I hope not on a 100Mbps). For example I said I get 250-300Mbps copying files, well I actually have closer to 500Mbps, shown with Qcheck.

[steingra]

Originally Posted by stanger89

I'd be reluctant to think that the fact you made your cables is causing problems at 100Mbps. I run all "home-made" cables and I get about 250-300Mbps on my GB-E connection, and I think that's limited by the speed of my array.

You could try Ixia Qcheck to see what your throughput actually is. The problems with the methods you've been using (file copy) is that you could be limited by your HDD read/write spead (though I hope not on a 100Mbps). For example I said I get 250-300Mbps copying files, well I actually have closer to 500Mbps, shown with Qcheck.

Well I just tried it with a store bought 350Mhz enhanced cable and the same thing happens. (I dont think I mentioned this before, but the racconworks SpeedTest program died when I tried this test on this machine) I think I might know why.

I was using the Full duplex 100Mbps setting. And when I start up the SpeedTest program, the cpu usage shoots up to 100% WOWWW! I cant believe its using that much CPU to do this.

Anyway, I changed the setting to half duplex, and now the Speedtest program doesnt die. And it is showing 80Mbps throughput. So its starting to make me think this poor old machine cant hack anything more than that. I am sure that much bandwidth/throughput is just fine. But the CPU usage concerns me, but I guess Sage TV client or any other media player is NOT going to use nearly that much bandwidth. So this should be a mute point. but it is interesting...

I dont really understand what the full implications are between full duplex and half duplex. Do you? It seems like in full duplex 100Mbps it makes the machine choke (if I am trying to do a massive file transfer over the NIC).

I am attaching a screenshot of NIC settings. I am wondering if the last 2 options are telling the OS to offload some of the RX/TX stuff to the NIC card hardware and therefore CPU wont be as strained???

PS
I did get the Qcheck program, so I will be trying that now too!!!! It looks pretty simple. Any suggested tests I should run/specific settings to try?

[stanger89]

Originally Posted by steingra

I dont really understand what the full implications are between full duplex and half duplex. Do you? It seems like in full duplex 100Mbps it makes the machine choke (if I am trying to do a massive file transfer over the NIC).

Full duplex means you can send and recieve simultaneously, while half-duplex you have to wait for one or the other to finish:

Full duplex:
SSSSSSSSSS
RRRRRRRRRR

Half duplex:
S S S S S 
 R R R R R

Probably not a big deal because you're probably doing mostly sending or receiving, not both simultaneously.

[teknubic]

Wow. So *this* is where all the network geeks are....
I'd just like to add my experiences to what Karen said:
Cables definitely do matter. Ethernet communication is very picky so the conduit should be treated carefully. For example, I've seen a kinked cable create 50% packet loss. Make sure your homemade cables are twisted right up to the RJ45 and are extra-smooshed. Also, stranded works best but solid will do.
Any good NIC will have a proc on board to handle the compiling/decompiling so your main CPU doesn't have to do the work. I can't think of one on the market that doesn't so you should be okay. You've been rebooting after changes to the network settings, right?
Jumbo packets and MTU sizes: I found that when Windoze boxes (2k and up) find out they're talking to one another, they up this size to 'talk faster'. This can be a PITA over a WAN, a bad link or especially through a firewall. There's a reg hack to make it stop doing that but I don't have it off the top of my head. You can find out if that's a problem by pinging the client with the '-l' option. e.g. C:\>ping -l 2500 192.168.1.10 type 'ping /?' to get the options and the explanations. Actually, if you're having problems with MTU on your home LAN, you got bigger issues or you're AR like me.
What are you doing that requires so much through-put anyway?

[steingra]

Originally Posted by teknubic

What are you doing that requires so much through-put anyway?

Well its not so much what I am doing, as just being a geek. Now that I am more aware of this situation (the fact that my NIC's and network wasnt running at 100Mbps like I *assumed* it was) I got sucked into trying to make it run that way

Which after some trial and error...I seem to be accomplishing a lot. The thing is, that I have a variety of older PC's laying around...and have been resurrecting them for the new life as HTPC clients. I like goofing around with each one, and trying to make it optimized as best as possible. And through it all, I have realized how certain pieces of hardware are working in ways that I never really paid much attention to before. Because it never mattered.

PC's are ok for day to day use, even programming, and you can take a lot for granted. But when it comes time to optimize slower pc's to make them work with streaming video over a LAN/WAN, all that goes out the Window. You have to scrutinize pretty much everything, one by one and get it right.

Tonight I have been playing some of my shows on the ~630 Mhz celeron PIII based sagetv client. Trying various settings. Once I got the network running as fast as possible without overloading the CPU...the sagetv client and VideoLan client are working fine without any skipping stuttering, etc. BTW, I did throw an inexpensive Geforce4 MX 4000 into it too. The 16MB matrox card didnt cut it. It was the least expensive 128 MB geforce card best buy had... And I loaded the Purevideo software on there, and am in process of experimenting with that. It seems to use slightly less CPU, than Intervideo, and less than SageTV decoders too...soo many things to try, too little time.

[Karen0302]

Originally Posted by teknubic

Jumbo packets and MTU sizes: I found that when Windoze boxes (2k and up) find out they're talking to one another, they up this size to 'talk faster'. This can be a PITA over a WAN, a bad link or especially through a firewall. There's a reg hack to make it stop doing that but I don't have it off the top of my head. You can find out if that's a problem by pinging the client with the '-l' option. e.g. C:\>ping -l 2500 192.168.1.10 type 'ping /?' to get the options and the explanations. Actually, if you're having problems with MTU on your home LAN, you got bigger issues or you're AR like me.

Now this is interesting. I have little experience with Win2K but quite a bit with XP and I have never seen TCP/IP running on XP create TCP/IP frames larger than 1500 (the TCP/IP maximum) all on its own. Additionally, even if it did, it would have problems communicating with other devices through switches and such UNLESS those switches supported Jumbo Frames ... and not many of them do.

So here is my question. Where did you see this happen? How did you detect it? And finally do you have any idea how to make it happen again? I want to do some experiments!

Karen

[teknubic]

Ah. I just figured it out again by discussing it with my coworkers and found I wasn't correct in that statement.
The two servers were communicating over a VPN tunnel. They would send out the 1500 byte packets and the VPN device would wrap them in it's encryption algorithm, making them larger than 1500 and would then drop them.
The solution was to either allow the VPN device to fragment the packets or to set the servers' MTU to ≤ 1472 (the 18 bytes were for the encryption).
So it appeared that the two servers were increasing the size of the packet but it was really the VPN. Sorry to get your hopes up.

[steingra]

Well I have been doing more testing using a wireless 54Mbps connection, and results still are not good enough.

Geeeze I cant believe the wireless connection is so crappy (at least when it comes to streaming videos). I must be doing something wrong, or overlooking some other things I can change. Im not giving up yet...on wifi dreams with sagetv/videolan clients.

I have used wireless for almost 2 years to connect to my companies servers using VPN connection. And used it for emailing, web browsing, watching occasional streaming videos, downloading files...and its all worked awesome. Never thought there was a problem with wireless until I started trying to really push it.

[stanger89]

Originally Posted by steingra

Geeeze I cant believe the wireless connection is so crappy (at least when it comes to streaming videos). I must be doing something wrong, or overlooking some other things I can change. Im not giving up yet...on wifi dreams with sagetv/videolan clients.

I have used wireless for almost 2 years to connect to my companies servers using VPN connection. And used it for emailing, web browsing, watching occasional streaming videos, downloading files...and its all worked awesome. Never thought there was a problem with wireless until I started trying to really push it.

Streaming video, especially full D1 quality video, is very demanding, much more so than most other applications. For example, wireless doesn't seem problematic for email, web browsing, downloading, etc because those activities are not sensitive to small changes in bandwidth. The wireless stream can drop for a second or two and you won't notice a blip with those activities. Similarly with streaming internet video, that is architected for the inherently unpredictable nature of internet transfers. That means it's buffered a LOT, often many seconds or up to most of the clip. Not that 'streaming video' is often at a relatively low bitrate to fit down average people's pipes.

What you stream with Sage is an entirely different animal, it's easilly 10x more data than even the highest quality streams, and due to the high data rates (often > 1/2 your actuall bandwidth) very sensitive to otherwise subtle changes in wireless connection strength.

One thing you might want to try is to adjust the number/size of buffers Sage uses for playback, a search should get you a few hits on that.

[steingra]

Originally Posted by stanger89

...
What you stream with Sage is an entirely different animal, it's easilly 10x more data than even the highest quality streams, and due to the high data rates (often > 1/2 your actuall bandwidth) very sensitive to otherwise subtle changes in wireless connection strength.

One thing you might want to try is to adjust the number/size of buffers Sage uses for playback, a search should get you a few hits on that.

I didnt know it was transferring that much more data, but seeing how it doesnt perform very well I can believe it. I will have to dbl check if I changed that registry key for buffering (on my laptop) I cant remember, its getting late

I KNOW I CHANGED IT ON AT LEAST 2 pc's just done remember if I did it on laptop.

I think I have seen the packet stream drop off to 0% on the wireless connection, then it will zoom back up to normal...for a while, then down, then gone, then back up. Its crazyyyyy Like you said, for pretty much all other applications you dont even notice it. Just hard core video streams.

I am sort of surprised (since you mentioned buffers) that sagetv cannot buffer a lot more data, even if its a settings in the properties file. Like for people who want to use wireless. It seems logical to me that if you were to buffer enough data on the client, that you could avoid almost all wireless dropouts...? no?

I dont know how much of a buffer it would take...and if that buffer would have to be an in memory buffer or if it could be a psuedo in memory and on disk combination buffer, depending on how much RAM you have free, and other params set up by user....

[stanger89]

Originally Posted by steingra

I am sort of surprised (since you mentioned buffers) that sagetv cannot buffer a lot more data, even if its a settings in the properties file.

Oh it can, when I was playing trying to get some HD playback to work (a while ago), I think I had it up to almost 512MB buffer, didn't help at the time .