Folks,

I have a Gb managed dell 5340 that seems to be choking on larger packets. While the ping command offers datagram size up to 65535, and I have used that size (usually use 65000 instead). Pinging thru the switch, from a cisco ASA to a NAS server, it starts having trouble around 30000 (60 to 100% loss, if I do it several times or if I set repeat to 100 I get 83%). At 40000 I get no replies, 100% loss

Anyone know what the normal frame size is?

Are there any negative repercussions I should be aware of before enabling "Jumbo Frames" on the switch? And is this likely to help?

The functional problem is tape backup is running about 5% expected speed when going from the NAS through the switch to the backup server through the scsi to tape it's slow. From the backup server through the scsi to tape it's fast (useless to us, but done as a test. We are trying to backup a 3.5 GB iso image and a 6GB folder of misc. files, seperately, just as a test.

TIA,

knudsen

Try testing with a crossover cable to eliminate the switch as the culprit.

That is about the only thing I haven't tired. My understanding is that would be a plug in and go thing. The old network settings should work as long as they were on the same VLAN?

I doubt your switch is the problem. Keep in mind that ping (ICMP) is a layer 3 protocol and the maximum packet size is 65535 bytes. Changing the size of your ping packet won't affect the way your switch handles frames since the switch operates at layer 2. I would skip enabling jumbo frames on your switch.

Do you have Gbit NICS on your NAS and backup server? The reason I ask is because if you have 100Mbit NICs on your servers they may be the bottleneck. Depending on what tape technology you are using the tape drive may be able to write faster than your NICs can pass data.

Most nics have a frame size of 1500 bytes, some will default to 1480 or 1425.

Run wireshark on your pc, and capture on your primary nic, then ping with, say a 10k size and see what the captured frame is.

Thanks guys!

All NICs are 1Gb as well as the switches interfaces. The tape drive is said to run up to 24 MB/sec at 2:1 compression. The manual says it should do a 20 GB backup in 13.5 minutes. I used a test file of about 3.4GB (an iso image). It took ~10 minutes to copy from my MP on a different VLAN, through an ASA5100 (100 Mb interfaces) to the NAS. It almost wouldn't even copy from the NAS to the backup server (same switch, same VLAN, no ASA). Past experiance with Dell servers tought me the NICs can get flaky. At least NIC GB1. So I swapped in GB2 (both integrated) and was able to copy the file in less than 8 minutes, maybe about 5. So we tried test backups of the iso image and locally from the backup server, the one with a scsi card connected to teh tape drive, it works wonderfully. From the NAS we got 2MB/sec, same as before and same as if we do real data backups Oh, and while testing the iso image, I use no compression.

Then I came back to work and just now read your replies.

Oh, ya, the 2MB per sec is always exactly 2.00 MB, never like 1.98 or 2.01 MB or anything. Weird. That makes me think it's a software limit somewhere.

Got it:

re: http://seer.entsupport.symantec.com/docs/290098.htm

Much easier reading than the symantec referenced KB article is in KB 912222. There is a table about halfway down the page that shows what combinations of TCP Chimney, RSS and NetDMA will work together. The problem is that 2003 SP2 and perhaps other updates turn them all on at once: http://support.microsoft.com/kb/912222/

Thanks for the input.