MTS

Gigabit Network Advice

13 posts in this topic

Hi folks - I'm having performance problems with my gigabit network and wondered if anyone had any advice on what to upgrade.

I've currently got this 8 port switch, just a cheapy and quite old now.

http://www.dynamode.net/giga_switches/SWG80010-R.htm

All 8 Network cards are either built into the motherboards, or cheap £30 cards by Dynamode, Genius and Dlink. All cables are cheap Cat5 (should I upgrade to Cat6?)

I'm not sure of the bandwidth I'm currently getting so if anyone could recommend a shareware util to measure that'd be much appreciated. I'm using this setup for streaming audio over the Lan from 7 slave machines into 1 host and it just isn't cutting it at the moment.

Thanks in advance,

Ian

Share this post


Link to post
Share on other sites

No, dont upgrade to cat 6, cat 5e should be fine. I dont know how good your 'cheap cat 5' is but if its in the walls, it would a major PITA to upgrade the cables.

I would upgrade the switch, since you say the switch is an older cheap switch, it likely doesnt support jumbo frames... I would upgrade to a switch that supports jumbo frames, before the cables....

Share this post


Link to post
Share on other sites

No - the problems are definitely network related. It's specialist audio software (Gigastudio and FX Teleport). I don't want to bore you with the details, but I have the option to run the 7 slave machines either directly into a mixing desk cutting out the LAN (with no issues whatsoever), or stream them back over the LAN into 1 PC which is where the problems begin (mainly a high latency - the software detects the speed of the network and sets the latency accordingly - a fast LAN would be a low latency but mine's very high).

Does anyone know of a shareware benchmark util I can measure my bandwidth?

My cables are all 2m long as the PCs are all racked up and cables straight into the switch so it'd be easy to change. I'm not sure if they're cat5e or just cat5.

No I don't think the switch supports jumbo frames. If I upgraded the switch to one which did, would I also need to upgrade the PCI network cards, or does this just apply to the switch?

Cheers,

Ian

Share this post


Link to post
Share on other sites

Hi,

you can use NTttcp and IPerf -- freeware tools measuring CPU usage and network throughoutput. If you'll fail finding them on the net -- lemme know I can send them to you (tiny console apps, plus I have a bunch of sample scripts to run them -- documenation itself is a bit nasty). To improve the speed I'd recommend:

1) play with Jumbo frame size (9K works fine for me and 16K results worser results -- f.e.)

2) put optimization settings into registry (TcpWindowSize set to 20M at least).

And PCI 32-bit/33MHz is definitley not enough for good performance. Either direct-plugged network cards (like CSA on i875/865 and "kno-knows-what" on nVida nForce 3 250) or PCI 64-bit/66MHz should be used. Replace is possible.

Anton

Share this post


Link to post
Share on other sites

That's interesting to know. These are 24bit, and each of the 7 PCs could be streaming upto 8 stereo streams, together with 448 virtual midi channels also on the lan so I guess gigabit would be best.

KOOLER - thanks - found IPerf so will let you know what I get tomorrow.

Cheers,

Ian

Share this post


Link to post
Share on other sites

Yes - comparitively to wav but it all adds up. Plus 7 * 8 stereo streams = 112 streams at 24bit so I guess that's close or over the limit on a 100kbit lan.

Ian

Share this post


Link to post
Share on other sites

It's not so much the bandwidth required but getting all the required data for each period time back to the desk. The main problem is switching latency and fitting the parallel streams into the single port towards the mixing console.

Slow 10/100 switches are typically in the range of 60 to 100usec.

In gigabit mode the faster switches can get this to below 10usec.

The mixing desk would have to support gigabit for it to make use of it and take advantage of the lower switch latency.

More avaiable bandwidth means less latency. Use the net tools to check current latency and max. bandwidth between two computers connected to your current switch 1) with nothing else connected 2) while audio gear is connected and running. What latency value is the audio gear setting ?

Jumbo frames will not help speed up any transfer between a Gigabit device and several 100Mbps devices. Jumbo frames does help between devices that have all participating devices (source,switch,destination) supporting jumbo frame and running with gigiabit ethernet. The gigabit switches may be able to fragment the jumbo frames into several smaller fragments for slower links with unknown? impact on performance.

How much does the latency issue affect useability of the setup ?

You may have to contact the manufactures to find out the information you need about latency, bandwidth, performance of your audio gear and possible switches.

Share this post


Link to post
Share on other sites

IF you do a search on SR for gigabit I think you should find a recent discussion about my attempt to get more network speed out of an upgrade from fast ethernet to gigabit ethernet.

To condense - there are a lot of other factors which seem to come into play when you go to gigabit, namely if the gigabit NIC is on the PCI bus, whether you have a Raid 0 drive, CPU power, etc.

It seems that gigabit will usually max out some other path of the data (ie HD throughput) before you max out the speed of the ethernet.

So upgrading your present network might not give you anymore throughput if the bottleneck is actually in another subsystem.

If you already have gigabit I think upgrading to jumbo might be the only option and I can tell you to see any significant increase your machines should all have Raid 0 & CSA NICs and fairly recent CPUs. So if not you may want to start investing in other hardware.

Share this post


Link to post
Share on other sites

I doubt that any switch could efficiently switch 100Mbps to a single 100Mbps port.

56 channels of 24bit/48KHz audio means 64'512 kbps then add packet overheads. To improve bandwith efficiency it may group 4 samples of 16 channels into one packet from each of the 7 devices. 48kHz becomes 42'000 packets each with at least 160bits of header = 6'720 kbps. This is about 8.5MBps which is probably more than what most 100Mbps network setup can cope with when not 'transfering large file one-to-one'.

4 samples @ 48kHz will take at least 84us delay to obtain. 60us for switch, <1us for cable, 20us for network interfaces, 2x10us for transmit and switch retransmit, plus another 15us for simple processing.

Total = 200us. All 7 mixers are synchronised so that the packets from all 7 appear on the input of the swtich at close to the same moment in time. Another 6 streams could delay the seventh by another ~360us. (The value needs to be checked by real-case testing).

I need to check the latency for my system at home which may be calculated in milliseconds (ms) for the whole system taking into account the MS Windows OS multitasking (switching) overheads. There are many real-time OS that allow much lower taskswitching delay (1ms time slice?) than Microsoft OS's (typically 10ms time slices).

Data compression can reduce the amount of data but larger sample buffers still mean larger delays for aquisiton and decompression.

AFAIK, MIDI data is 31.25Kbps. But I don't know if this covers the usuall 16 channels or not ?? 2kbps x 448 = 896kbps.

Share this post


Link to post
Share on other sites
I doubt that any switch could efficiently switch 100Mbps to a single 100Mbps port.

Why?

56 channels of 24bit/48KHz audio means 64'512 kbps then add packet overheads. To improve bandwith efficiency it may group 4 samples of 16 channels into one packet from each of the 7 devices. 48kHz becomes 42'000 packets each with at least 160bits of header = 6'720 kbps. This is about 8.5MBps which is probably more than what most 100Mbps network setup can cope with when not 'transfering large file one-to-one'.

Well, 7 100 mbit ports and one gbit port would be enough and that's a big difference with 8 gbit ports.

I need to check the latency for my system at home which may be calculated in milliseconds (ms) for the whole system taking into account the MS Windows OS multitasking (switching) overheads. There are many real-time OS that allow much lower taskswitching delay (1ms time slice?) than Microsoft OS's (typically 10ms time slices).

Data compression can reduce the amount of data but larger sample buffers still mean larger delays for aquisiton and decompression.

AFAIK, MIDI data is 31.25Kbps. But I don't know if this covers the usuall 16 channels or not ?? 2kbps x 448 = 896kbps.

Doesn't the time slice only matter if there's another process with high CPU usage?

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!


Register a new account

Sign in

Already have an account? Sign in here.


Sign In Now