ftp of flat files takes longer on one server

[James Ellis_1]

We have two servers, one is an N class and another is an L class server. Both use fiber channel cards.

N Class: 8gb RAM, directory resides on HP SureStore RAID, 4 RISC processors and each has 440 mhz.

L Class: 8gb RAM, directory is on a SAN, 4 RISC processors and each is 550mhz.

Developer has stated that the load process of FTPing the flat files to the N Class box takes 5 hours, whihc it takes 2.5 hours on the L Class box. The L Class box is a production box.

Several questions come up. First is, I highly doubt that the L Class with 550mhz would make that much difference, at least not cutting the time to FTP the files to 2.5 hours.

Does the RAID array cause the system to slow down because of striping? Would these two storage device, HP RAID and SAN make that much difference?

Any input on what to compare and what to benchmark to get to the root of the problem. The load starts at 3:00am, so its not like a whole bunch of process are running at that time.

Thanks.

[Steven E. Protter]

System load can have an effect like this but not so dramatic.

lanadmin -x 1

1 is the lan number from a lanscan.

You might find your duplex settings are wrong on the slower machine.

To insure proper duplex, set up a /etc/rc.config.d/hpbtlanconf file with settings. I'll attach an example.

Its also important that on 100 BT and slower LAN cards that the switch settings bet explicit, not Autonegotiate. Very important on Cisco switches.

SEP

[A. Clay Stephenson]

This is almost certainly a difference in network settings. Unless this is Gigabit, you really should hard set the speed settings on BOTH ends of each connection. If you want to eliminate disk i/o as a possible difference then do something like this:
timex dd if/dev/zero bs=64k count=160 of=temp1


That will write a 10MB file.

[James Ellis_1]

Here is my info on lanadmin -x 1:

Speed = 1000 Full-Duplex
Autonegation = ON

[James Ellis_1]

the interesting thing is on the L class server the setting is as:

10 Half-Duplex Autonegotion-ON


You would think a 1000 Full-Duplex like on the N Class box would provide faster throughput.

Thanks.

[John Dvorchak]

James, having your NIC setting different can really affect performance. The key is having the NICs set the same as the switch port. It is well known that most NICs and switch ports don't do well when they are set to "autonegotiate". I would suggest that you set it to 100FD with lanadmin:

lanadmin -S 100FD x

Where x is the PPA number from lanscan for that adapter. Be sure and check with the network folks before you change this to assure that the switch port is set to 100FD also.

[Steven E. Protter]

James, you have your answer.

You need to get the switch checked.

You can do a lanadmin -X 100FD 1 to try and bump up the speed on the fly.

My /etc/rc.config.d/hpbtlanconf

suggestion will solve the problem permenently so long as the switch settings are not auto negotiate.

A solution lies in the thread.

Good luck and let us know how it turns out.

SEP

[James A. Donovan]

James,

If your positive that it's the L-class that has the faster ftp transfers, yet the NIC card on that box is a factor of 200 slower, then your problem is almost certainly the disk.

Is the N-class attached to an HP AutoRAID 12H?? Also known as the "slowest disk array under the sun".

There could be features to the filesystem on the N-class that could be causing additional overhead. Check the fstab files on each server.

Since the L-class is SAN attached, it's very possible that the backend disk (or disk array) supporting the L is far superior than what you have attached to the N. Disk array caching can make a big difference.