10baseT speed improvements?

Posted by: chris_vd

10baseT speed improvements? - 01/09/2000 11:18

I have my Empeg plugged directly into a switch, here at the office. I shunted the specific port down from 100 to 10Mb, and left the other settings on autodetect. According to 3Com's handy switch info tools, Empeg, while having large packets, only uses a paltry 40% of the bandwidth. On top of that, the onboard port is set to half duplex.

Does anyone know a way to change the port to full duplex? This would nearly double the transfer rate to the Empeg, effectively cutting sync time in half. Is it hard wired to half duplex???

Posted by: altman

Re: 10baseT speed improvements? - 01/09/2000 14:06

I think the controller should be autodetecting full-duplex capability on the hub, though I've not checked this.

Setting it to fd would not impact the transfer times - the reason for the lack of full bandwidth usage is simply that the protocol running over TCP is the same one that runs over USB and serial - ie, it's an error correcting stop & wait (not windowed) protocol. ftp to the unit can get 1MB/sec sustained: we're planning on some protocol improvements which will help enormously with this.

Hugo


Posted by: Drakoz

Re: 10baseT speed improvements? - 02/09/2000 01:57

Enabling full duplex would not noticeably speed up transfers to your Empeg. Full duplex only means it can send packets to and from the unit at the *same* time. But, when uploading music, the majority of data transfer is to the Empeg with only a very minimal amount of date from the unit. So, full duplex doesn't really matter for an Empeg.

The 40% transfer utilization is disappointing. I hope that can be improved to full bandwidth. Some protocols are just inefficient, though, so I can understand.

Mike

Posted by: PaulWay

Re: 10baseT speed improvements? - 02/09/2000 15:58

Hugo, does this mean that the new version of the protocol will be windowed with serial and USB too? If USB is roughly the same basic data rate and latency this might speed it up too.

Always liked windowed protocols...

Save the whales. Feed the hungry. Free the mallocs.
Posted by: altman

Re: 10baseT speed improvements? - 03/09/2000 03:14

It may be, other possibilities include a separate socket connection for data (like ftp). USB has other limits like interrupt latency, as it interrupts for every 32 bytes of data (mk1) or 64 bytes (mk2). The protocol gets nearer maximum possible throughput on USB than on ethernet - max USB throughput is limited by the speed we can access the USB slave controller device (which isn't fast - the chip's design won't allow it :( ).

Hugo