ISCSI LUNs not reachable after Lan Card set to 100FD

[Lucy2009_1]

Hello,
Hope someone out there has the answer for me.
Here is the problem:
I set the iscsi lan card to be 100FD in /etc/rc.config/hpietherconf. System came up without any error.
# lanadmin -s -x 3
Speed = 100000000
Current Config = 100 Full-Duplex MANUAL

On the CISCO swith port, it is set to 100FD manual as well. Other system can ping this NIC address. But all the ISCSI luns did not mount after system came up. I restarted iscsi daemon and it complained not finding ISCSI disks. From the system, it can ping the storage. What can be the problem? The ISCSI Nic was on 100 half-duplex/auto. But the swith port was on 100Full-duplex/mannual. Users complained slow speed, that's why I changed the nic speed to match the switch port setting. Any idea what's wrong? By the way, this happened to two systems running hpux 11.23 and 11.31.

Thanks,
Lucey

[Jeff_Traigle · Accepted solution]

I didn't even think the driver would allow you to manually set the speed for a Gigabit Ethernet NIC. Never worked when I tried it in the past anyway. They are supposed to be allowed to autonegotiate. Have you tried setting the switch port to autonegotiate to match the server's previous configuration?

Don't know that it'll help the iSCSI problem. Presumably, if you've got TCP/IP connectivity, it should work. Haven't used it myself so don't know of any possible caveats.

[rick jones]

Check your link-level stats with lanadmin to make absolutely certain you didn't introduce a duplex mismatch somwhere by forcing duplex rather than using autoneg. Ping on an otherwise unloaded interface will *NOT* uncover a duplex mis-match because it does not cause simultaneous traffic.

[Lucy2009_1]

Thanks all for the input. The server side that I checked using "lanadmin -s -x " showing 100FD, atuo off. This is set by the startup file /etc/rc.config.d/hpietherconf at boot up. The switch port side also manually set to 100FD, auto off.

Anyway, I changed switch port to auto and removed setting in hpietherconf to back to auto. Rebooted the system and it came up as 100 half-duplex. ISCSI luns were all mounted.

I just curious what else to look at? Leave it as 100 HD?

Thanks,

Lucey

[Bill Hassell]

When you see half-duplex, you are guaranteed that auto-negotiation has failed. And yes, it will be very slow in this mode. lanadmin -g 3 will show you a bunch errors (in the second half of the listing) which won't happen in full duplex mode. 1Gb (and 10Gb) cards must always be set to auto as well as the switch port.

[Lucy2009_1]

After reading Rick's attachment, I see that 1000Base-T will have to use autoneg. One of these is 1000Base-T so I think we have to use autoneg.

The other one is 100Base-T. Why the manual set did not work on both server and switch port remains mystery.

Lucey

[rick jones]

I think that may be lanadmin -g mibstats 3, but still, yes, check the link-level statistics.

Again, just in case that was not clear, check the link-level statistics :)

IMO, one should never force duplex unless one has equipment with a known autoneg problem and said equipment absolutely cannot be replaced. But you should have already seen that in the attachment in my previous posting :)

[Lucy2009_1]

Thanks Bill and Rick. Here is the output of the lanadmim. The server is L2000 running hpux 11.11. Both sides are using autoneg now in order to see iscsi luns.

#lanadmin -g mibstats 3

LAN INTERFACE STATUS DISPLAY
Wed, Jan 12,2011 16:56:19

PPA Number = 3
Description = lan3 HP A3738A PCI 10/100Base-TX Ultimate Combo [100BASE-TX,HD,
Type (value) = ethernet-csmacd(6)
MTU Size = 1500
Speed = 100000000
Station Address = 0x108327c508
Administration Status (value) = up(1)
Operation Status (value) = up(1)
Last Change = 744
Inbound Octets = 7139348
Inbound Unicast Packets = 8831
Inbound Non-Unicast Packets = 2928
Inbound Discards = 0
Inbound Errors = 0
Inbound Unknown Protocols = 9
Outbound Octets = 9287698
Outbound Unicast Packets = 10067
Outbound Non-Unicast Packets = 7
Outbound Discards = 0
Outbound Errors = 0
Outbound Queue Length = 0
Specific = 655367

Ethernet-like Statistics Group

Index = 4
Alignment Errors = 0
FCS Errors = 0
Single Collision Frames = 277
Multiple Collision Frames = 36
Deferred Transmissions = 1042
Late Collisions = 462
Excessive Collisions = 0
Internal MAC Transmit Errors = 0
Carrier Sense Errors = 0
Frames Too Long = 0
Internal MAC Receive Errors = 0

[rick jones]

So, per the previous attachment, the late collisions are a strong suggestion you have a duplex mismatch. Given that you say you have put the NIC back to autoneg, that suggests that either the switch port is still hardcoded (perhaps the wrong switch port was reconfigured?) or you have a switch which has broken autoneg. I don't know how large a switch you have, but given how 1 Gigabit/s capable switches (which, presumably should autoneg just fine since it is a required part of the GbE standard) have come-down in price, if your's is an old, 100BT only switch, it might be a Very Good Idea to replace it with something rather newer. Naturally, I'd prefer that you start at http://www.hp.com/go/networking :)

[Lucy2009_1]

Rick,
The switch is CISCO Catalyst 3550. 100BT ports. If I move to another CISCO switch with 1GB port, will autoneg on both end work correctly, nic is 10/100 Base-T?