Slow file transfer

[Rudy Williams]

I have two HP-9000 servers (rp2430 and 5430) both with: one 100Base-T network connection and one 1000Base-F network connection. All four NICs are confiugred on the same subnet.

An FTP to an NT server system is very fast, as expected. However, when doing a file transfer from one ot the HP-UX boxes to another, the file transfer is horribly slow. As in painful, fingernails on chalkboard slow. A 1 MB file can take a minute. The odd thing is, this is also true if ensuring that we are using the gigabit NIC's in the two machines for the transfers. (We FTP to the IP address configured on the gig network card.)

No network errors appear in the syslog.log files.

Any ideas?

[Bill McNAMARA_1]

is it just ftp?
try rcp
just to amuse me..

Sounds like you need a patch..

lanadmin can possibly help identify any protocol errors..

Later,
Bill

[V. V. Ravi Kumar_1]

hi,

check ur duplex settings on the interface card and on the switch, try to put full duplex, ftp largely depends on duplex settings.

and also is there any problem with telnet becoming slow?

regds

[Martin Johnson]

Try doing a "traceroute" between the two servers. This will tell you the number of hops and the time it took to get there.

Do a "lanscan" to verify your lan cards are up and supported. "lanadmin" can be used to display the counters. Check for excessive errors.

Also check the speed of the cards and the IP addresses. The 100Base-T should be talking to the 100Base-T on the other system. Same with the 1000Base-F.


HTH
Marty

[Bill Hassell · Accepted solution]

Use lanadmin -x to display the speed, duplex setting and the state of autonegotiation. For 100Mbit, a report of half-duplex means that autonegotiation has failed and you'll now see collisions and FCS errors reported in the lanadmin stats (use lanadmin interactively--you'll need the LAN instance numbers from lanscan). The ppa (Physical Point of Attachment) value in lanadmin really means LAN instance number for 11.xx opsystems.

If half duplex is the setting, you probably have an unlucky cable length between your machine and the switch and timing has misconfigured the duplex setting. To fix this temporarily, use lanadmin -X to set the 100BaseT cards to full-duplex. (Also change the switch to manual/100Mbit full-duplex) Do not turn off autonegotiation for the Gigabit cards as this is a requirement for the protocol.

[Rudy Williams]

The lan0 NIC is 100-BT.

lanadmin -s reported 100 Mb/s
lanadmin -x reported HD
A-ha!

lanadmin -X 100FD 0 solved the problem!

Now, a side question. I have 4 11i servers that I want to communicate with each other via the gig Ethernet. I notice in the syslog.log file that the source IP is on the 100-BT card, not the gig card.

Recall both 100-BT and gig cards are on the same subnet. Is there some way that I can force the gig NIC to be used when communicating between the HP-UX boxes?

[rick jones]

configuring multiple NICs into the same subnet is il-advised. without further work on your part, traffic will be recieved on either NIC, but only one will be used for outbound traffic (either the first or last ifconfig'd).

if you want one link to be a fail-over for the other, you might look-into the LAN Monitor functionality of Auto Port Aggregation.

If the number of IP's involved is small and there are no fail-over concerns, then you can configure explicit host routes for the remote destinations, pointing at one or the other of the local interface IPs. This will work for local traffic, it will rely on proxy ARP for remote IPs.

If you use ndd to set ip_strong_es_model to one (1), then traffic will exit the system on the interface on which it entered - so from the server standpoint that is cool as the incoming request picks the IP address. for initiation, i'm honestly not sure what happens - I've never tried it there.

you would still have to use explicit IP's in the calls to ftp etc to get that remote IP used unless you have separate names for each IP.

fwiw, it is ok to have multiple address aliases in the same subnet on a single NIC (single NIC as far as the transport can tell - IE an APA aggregate is OK too). it is also OK to have multiple IP subnets running on the same switch.

[Yogeeraj_1]

Hi,

I would consider doing the same thing using HP Auto Port Aggregation (APA).
APA is defined as allowing the exact same IP address to exist on two different LAN cards at the same time, for redundancy as well as increased throughput.

However, you cannot use ordinary hubs or switches with APA as the switch must support the APA protocol.

In order for an APA link to form, there must not be ANY IP addresses ifconfig'd on the lan NIC's. Otherwise you'll see an error and the lan aggregate will not form.


If you presently have lan0 in your netconf, just change that to lan900, which will be the lan formed by APA. Of course make sure you get the rest of the hp_apaconf and hp_apaportconf files set up too, per the manual. Its really pretty easy once you do it a couple of times.

On my system what I did was to remote the IP addresses and unplumb the lan cards being used in the aggregate, then edited the config files, and ran /sbin/init.d/hpapa start, and verified the lan aggegate was up e.g. with lanscan -q and lanadmin -x -v 900. After that, ifconfig lan900 inet 192.168.1.1 up and it came up and ran fine.


Have a look at the following document for more information on how to do this.
http://docs.hp.com/hpux/pdf/J4240-90016.pdf

Hope this helps!
Best Regards
Yogeeraj