FTP speed on blade servers

[James George_1]

Hi

Iam new in Linux.

This is my problem. We have 9 servers in a blade enclosure , all running Linux , Redhat.
The ftp is very slow between two of these servers. We have 2 subnets invloved here. One of of the servers is on 10.17.3.X and all the other servers on 10.17.13.X . FTP ( 4 GB file) between servers on the same subnet is done in 2 minutes. But from servers on one subnet to the other subnet server, it takes hours !! . WE have involved HP and our n/w team and everything is ok on the n/w side / router / switch side. The Speed setting is also (correct 1GB FULL DUPLEX ) and we don't find any problem.

Now we are doubting the OS side ...Can some one help ?

Rgds/ James

[Rob Leadbeater]

Hi James,

I doubt that this is an OS thing. Sounds more like something in the network to me...

How are the two subnets linked together ?

Cheers,

Rob

P.S. You might want to let people know exactly which version of RedHat you're running just in case...

[Ivan Ferreira]

>>> WE have involved HP and our n/w team and everything is ok on the n/w side / router / switch side.

How did you determine this?

Can you post the output of netstat -ni of both servers?

Have you tried a packet trace/dump, for example, with tcpdump?

[James George_1]

Here is the version details .

2.6.9-42.0.8.ELsmp #1 SMP Tue Jan 23 12:49:51 EST 2007 x86_64 x86_64 x86_64 GNU/Linux

We even did an ftp from outside the blade to the blade server and it went very well. Looks like it got something to do with the switch inside the Blade Enclosure. But , none of us are good in Linux / blade which is a new arrival in our environment.

Rgds/ JAmes

[James George_1]

Yes, we did the ( HP) tcpdump . We tried ftp between different servers in and out of the different subnets . The problem is only on this specific server, rest all the servers are transfering data very fast.

rgds/ James

[James George_1]

Hi Ivan

here is the netstat -in outputs

mixdb01p:/root>netstat -in
Kernel Interface table
Iface MTU Met RX-OK RX-ERR RX-DRP RX-OVR TX-OK TX-ERR TX-DRP TX-OVR Flg
eth0 1500 0 2743481325 0 0 0 4544026452 0 0 0 BMRU
lo 16436 0 61277474 0 0 0 61277474 0 0 0 LRU





mixdb02d:/root>netstat -in
Kernel Interface table
Iface MTU Met RX-OK RX-ERR RX-DRP RX-OVR TX-OK TX-ERR TX-DRP TX-OVR Flg
eth2 1500 0 376081554 0 0 0 785191661 0 0 0 BMRU
lo 16436 0 20926155 0 0 0 20926155 0 0 0 LRU


James

[Stuart Browne]

How big is the subnet mask on each end?

You say you're using different networks, so unless they've got a very large subnet mask, the FTP has to be going through a router somewhere, which means it's probably leaving the blade enclosure and coming back in.

[James George_1]

The subnet mask are 255.255.255.192 and 255.255.255.128.

You are right, there is cisco 3020 switch , part of the blade .

Rgds/ James

[Stuart Browne]

That's a Layer 2 switch, not a router, so it's leaving the 3020, hitting something outside the enclosure, then coming back in.

Well, that's what I feel from what I've read.

[Stuart Browne]

Forgot to add, 'traceroute' is your friend.

[James George_1]

traceroute shows 1 hop and its fine.

rgds/ James

[Stuart Browne]

Then there's still some information missing, given they are both on different subnets, and both subnet masks are fairly tight.

Show routing tables on both.

Also show iptbales rules on both (both -t filter and -t nat).

[Rob Leadbeater]

Another thing to consider is that blades tend to have a number of NICs...

Please post the output of

# ifconfig -a
and
# netstat -rn

from a server on each subnet. As others have said, this would appear to be a routing issue.

You've said that Linux is new to you, so it would be useful to see actual output of commands, rather than you just saying the output is "fine".

Cheers,

Rob

[Matti_Kurkela]

With blade servers (at least the new c-series), the speed/duplex state you're seeing with the OS tools (ethtool and the like) will refer to the blade rack's internal wiring only. The internal wiring is forced to 1GB Full Duplex at the hardware level, so it would be surprising if this is incorrect.

To know the speed/duplex settings of the physical outgoing interfaces, you must examine the settings of the enclosure's integrated switch module. If using a pass-through module instead of the integrated switch, the external switch *must* be 1GB Full Duplex capable: the pass-through module is 1GB Full Duplex *only*.

MK

[Steven E. Protter]

Shalom,

If it were all systems on the blade, I'd suspect you outstripped the blades network capacity with demand.

I would suspect in this instance that its a network switch port issue, perhaps duplex is set manually and it should not be.

SEP