Cisco Support Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Announcements

Welcome to Cisco Support Community. We would love to have your feedback.

For an introduction to the new site, click here. If you'd prefer to explore, try our test area to get started. And see here for current known issues.

New Member

Rx and Tx speeds differ by 300%

I need a discussion on the settings of a NIC card that could account for such a difference. I have a PC that's attached to a Cisco 3750 switch, at 1GB / Full-duplex. When I copy files from the PC, to a NAS (on same 3750 swtich), I get ~87Mbit/sec with an 8.7GB test file. Now when I reverse the direction, (same file, same machines, all things equal), I get ~320Mbit/sec. The switch port has all defaults set; nothing special, and *no* CRC/runts/giants/collisons. I believe this has to do with the NIC card, and buffering settings(?) The support org covering the PC is useless, so I'm coming to the Cisco NetPro forums where the *real* experts are. Please give all ideas! I don't really need Cisco-specific support, I need support on the concepts of buffering & flow-control.

4 REPLIES

Re: Rx and Tx speeds differ by 300%

Here are two options to test:

1: Have you checked the TCP window size on the PC? Check what happens when you set it (in the registry) to a high value. Microsoft says it should not be relevant anymore as it is tuned automatically but I am not sure if they tested with such extremely large files as yours.

2: The limiting factor in this setup is likely to be the PC's harddrive. With 8,7G there is little chance that the file can be cached so it will have to be read from disk. This is a relativly time-consuming process, especially when the file is heavily fragmented. The read-buffer is emptied into the network and the harddrive likely needs a few rotations to fill it again. Writing to the disk is faster because the write-buffer is never empty when the data is delivered at such high speeds.

I suppose the NAS is far better equipped to serve large files. This may explain the difference. To verify this theory, you may want to repeat the test with a smaller file, one that can be cached by the PC before sending it.

Regards,

Leo

Re: Rx and Tx speeds differ by 300%

You are best using tools designed for testing throughput as the previous poster said the PC hardware is likely a limiting factor. I have used IPerf sucessfully in the past - this can generate streams of UDP & TCP at very high rates. It is still possible though that the PC hardware will be a limiting factor.

Start with a baseline using a cross-over cable between the 2 PC's and see what the maximum throughput you can achieve. Once you have this connect the PC's to the switch and repeat.

It may be worth getting hold of a dedicated tester (or renting?) such as SmartBits as this won't have the limitations of a PC.

HTH

Andy

New Member

Re: Rx and Tx speeds differ by 300%

The system originating the copy is an appliance of sorts, but it's a dual-Xeon 3Ghz system with an Adaptec 2200S RAID adaptor with 136GB Ultra320 HDs. It's quite a speedy machine actually.

Re: Rx and Tx speeds differ by 300%

But you are running Windows on it? In that case, I would still advise you to test with a hard-coded TCP window size.

Considering the disk option: Are you actually running a raid5 configuration on this magnificent controller? If you have a mirrored setup using only two disks, you might not need to seek any further.

Reagrds,

Leo

93
Views
0
Helpful
4
Replies