Network disconnects.

[noka]

LC2000 w/RAID-1M card, (5) 18GB hotswap drives (RAID-5 config), Win2K SP2 and also following security patches and hot-fixes (w2ksrp1.exe, Q314147.exe).

Switch is ProCurve 2324 (J4818A, unmanaged). All PC's set to "auto-detect (LC2000)" or "hardware default" (VL400) for speed and duplex setting, to create 100Mbps/full-duplex environment.

Problem: Application crashes (usually returns to Win2K desktop but sometimes hangs). Inspection of Win2K System Event log shows Event ID=4202,4201, source Tcpip sometimes one or two entries, sometimes numerous entries. Text is something like:

"The system detected that network adapter Connection was disconnected from the network, and the adapter's network configuration has been released. If the network adapter was not disconnected, this may indicate that it has malfunctioned. Please contact your vendor for updated drivers."

Not sure what driver we use (probably taken automatically from Win2K CD).

Strange thing is that clients (Vectra VL400) also exhibit same symptom sometimes.

Any idea?

[Arimo Laine_2 · Accepted solution]

Hi.

First of all, try to update your switch firmware. Current version is F.05.14, that can be downloaded from www.hp.com/rnd/software/switches.htm.

The upgrade instructions are in switch Management and Configuration Guide, Appendix A, File Transfers. The only available way is to use a console connection and XModem. Manual download page is www.hp.com/rnd/support/manuals/index.htm.

Another thing, get rid of Microsoft drivers and use the appropriate HP drivers for your server.

HTH,
Arimo

[noka]

Forgot one thing. Two ProCurve hubs (J3295A) are plugged into the switch (each to separate switch port) and VL400's are connected to hubs via patch panels. I know hubs only run at half-duplex but system should work, I think.

About f/w download to ProCurve switch:
1) How doyou check latest f/w rev?
2) How do you get X-modem connection and what cable is used?

[Arimo Laine_2]

Whoh.

As for the hubs, yes, the system should work. But if it's possible, maybe you could set the Vectra NICs to 10HDx (or 100HDx, cannot remember whether the hubs are 10 or 10/100 devices...) instead of default.

As for the FW update:

For console connection you need a null-modem cable between the switch and PC COM port. You can use Hyperterminal as a console client, but actually I'd advise to check out Tera Term Pro (http://hp.vector.co.jp/authors/VA002416/teraterm.html),
that one works with HP switches with default settings.

Once you have the console connection, you can use command show version to find out the current firmware.

HTH,
TAL

[noka]

Thanks. The HP hub manual says:

"These hubs are autosensing and autonegotiating devices that provide the flexibility of adjusting to different network connection speed settings. The hub ports
always operate at half-duplex."

Seems to me that server set to "auto-detect" should play w/switch. Switch should play w/hubs, and client PC's set to "hardware default" should play w/hubs.
No?

By the way, anyone know what setting of hardware default means for VL400 PC?
- Media Type choices are:
100Base TX
10Base TX
Auto Select
Hardware Default (using this one)
- Duplex Mode Choices are:
Full Duplex
Half Duplex
Hardware Default (using this one)

[Arimo Laine_2]

I've no clue what's the HW default on vectra NIC :-) If VL400 is even relatively modern, it's most propably 100FDx.

If you're connecting them to the hubs, I'd anyhow check if setting them to 10HDx solves the problem with the vectras.

TAL

[OLARU Dan]

Are you still having problems with your network disconnects?

[noka]

Yes, it is still a problem but for reasons too complicated to go into here, we did not try a switch firmware upgrade. We will do that soon. If that does not work, maybe we do some network monitoring, to try and trap some event, etc. Curious, do you have some theory we could test?

[OLARU Dan]

Noka,

1. You can chech the firmware revision of swt 2324 (J4818A) after you connect to it (via serial port) and choose something like "Status and Counters" and then "General System Information" (this is the case for 4000M - which I have)

2. The only version of the firmware (OS) for this swt is F.05.14, which HP did not change since 26 September 2002. It is available for download from http://www.hp.com/rnd/software/switches.htm - but you might already have this version installed in your swt (check it on the swt before downloading it)

3. If you can't connect to the swt from the first attempt, adjust the Baud Rate of transmission on the terminal emulator to minimum of 1200 baud and then try again. But do not try to transfer the firmware from your PC to the swt at this speed, since it will take ages (the OS has ~1.54 MB).

4. Let's say you have the latest OS on your swt.

4.1 Check the HUB-to-SWT uplinks are correct (you use cross-over cables if you don't have a special uplink port on each hub)

4.2. If all PCs that connect to one J3295A (let's say HUB1) have the symptoms you describe AND when you replace the hub with another one the symptoms dissapear, then HUB1 is broken and should be forwarded to HP for repairs.

4.3. If you can't blame any particular hub (many PCs connected to any hub, and PC that directly connect to the swt have the symptoms), then the swt might be faulty and may need repair.

5. On one of your PCs set the Media Type to 100BaseTX and Duplex Mode to Half Duplex. If the PC still has the symptoms, try 10BaseTX. Auto-negotiation on switches doesn't always work as advertised.

6. To see if the PCs have the correct NIC drivers and the networking is OK on PCs, connect 2 PCs through a good cross-over cable (without passing through the swt or hubs) and check if the two computers can see each other.

[noka]

Hi, thanks for the response. A few comments (sorry it is so long):

1/2. J4818A - yes, we can check f/w. I am reasonably sure it is less than F.05.14 but need to verify. J4818A switch is unmanaged and 9-pin d-sub connector is 'download port' (e.g. for f/w upgrade). I am not sure if it will respond to other commands you mentioned but can check.

3. I had some trouble connecting to switch in my lab (not system in question) but was able to successfully connect after adjusting a few parameters (e.g. error correction type).

4. Hub was replaced by switch - same problem. Patch cables were checked for correct type - replaced - same problem. Switch had already been replaced (unsure of f/w rev when that was done, however) - same problem.

5. We can try some other duplex/speed settings but HP doc specifically states that connection of any device to unmanaged switch (J4818A) requires "auto" setting, otherwise duplex mismatch occurs and will cause more comm problems. We have other systems working successfully with same h/w - but again, not sure about switch f/w rev.

6. Computers can see each other and same NIC drivers are used successfully on other similar systems we have installed elsewhere.

NOTE: We did find on HP web that F.05.14 switch f/w solves buffer overrun problem in switch due to network connects/disconnect transitions. System has some "moveable" PC's on carts. Users supposed to shutdown and move client to another room, reconnect, power up and reboot. We don't know if they shutdown completely, although they are advised to. In any case, we wonder if these shutdown/restart transitions might be causing switch buffer overrun because:

1. Just before server/system problem, we notice Event 4201 and 4202 in Win2000 System Event log (i.e. NIC connects and disconnects). This occurs maybe 10 sec before the problem. We theorize the buffer overrun causes switch to reset/reboot, which causes our server to lose DB connection, since it is done over network layer. This would cause our application to fail on the server with resultant application crash on clients. We not also that the PC's are sometimes found at the Win2000 desktop on all PC's, so it doesn't appear to crash the OS.

2. The problem is intermittent but seems to occur approx every 9-12 days.

So maybe the switch f/w upgrade will help, we will see.

[OLARU Dan]

If you have Intel NICs, and your J4818A doesn't have F.05.14, then you're in trouble. Also this f/w solves A LOT of half duplex problems. I attached the release notes for the f/w - maybe this will determine you to go install the new f/w.

The nasty thing is that you have to go near the switch and plug a serial cable from a laptop to the "Download Port" of the switch - since you can't reach it by telnet.

Today I received my 2 J4818A machines - I will experiment a little with them and then call you back.

[noka]

The PC's in question (VL400) have 3COM onboard NICs. We do have some other systems with VL420 and Intel NIC's, so thanks for the warning. We will try the switch firmware update anyway (soon I hope).

Our connectons are intended to run 100Mbps, full-duplex, so we set our server and PC NIC advanced properties to "auto", or in some cases "hardware default" (but I'm not sure exactly what that setting means). All this is specified by our R&D group.

[OLARU Dan]

OK!

1. I first had to solve the problem with the COM1 port on my HP Vectra. The port was blocked in BIOS by some security setting (check out the "COM Ports" topic that I posted on the Switches thread for the solution of hidden COM ports.

2. I managed to download the F.05.15 f/w to my J4818A machine. Previously it had version F.04.08 installed - so you REALLY need to do the f/w upgrade.

3. Here's how, if you're using a Windows laptop with COM1 port enabled:
3.1 Start >Programs >Accessories >Communications >HyperTerminal
3.2 Cancel the "Connection Description" window that pops up
3.3 Press the righmost button (Properties) on the toolbar of the window "New Connection - HyperTerminal". The Status bar of the active window says: Disconnected | Auto detect | Auto detect
3.4 "New Connection Properties" window pops up. On the "Connect using:" drop down list, select "COM1", then press "Configure..." button
3.5 "COM1 Properties" window pops up. Select "115200" on the "Bits per second" drop down list, select "None" on the "Flow control" drop down list. OK twice.
3.6 Press [Enter] twice on the HiperTerminal window. A copyright text should appear in maximum 5 seconds. If it does not, close the HyperTerminal on the laptop, unpower the switch (some users might phone you at this step), power it back again to reset it, and redo steps 3.1 to 3.6 until you get the copyright message.
3.7 The prompt "HP ProCurve Switch 2324#" appeared. press "?" to get the help on the 3 commands available.
3.8 Type "download" and press [Enter] twice
3.9 Mouse "Transfer/Send file..." on the Menu bar
3.10 Choose "Xmodem" in Protocol drop down and then browse for the Filename F_05_14.swi on your HDD. Press "Send" button.
3.11 In maximum 5 seconds you should have transmission numbers rolling on the new poped-up window.
3.12 I you don't, then you hit "Cancel", type "Exit" at the switch prompt, and redo steps 3.6 to 3.11 - ONLY if you are still reading this rather lenghty procedure.


PS: C'mon noka, assign me some points: I want to catch up with Ron and Arimo :-)))). You must be logged on the site to do that.

Cheers!

[OLARU Dan]

<>

1. Most probably you've setup the NICs' "Duplex Mode" to "Hardware Default" (usually, it also has "Full Duplex" and "Half Duplex" options - to avoid the autonegotiation of the duplex mode) and "Media Type" to "Auto Select" (usually, it also has "100BaseTx" and "10BaseTx" options - to avoid the autonegotiation of the link speed). But for the Autonegotiation to operate, you HAVE TO set the NICs and any managed switch to "Flow Control" = "Enabled". For J4818A, the flow control is enabled by default - don't forget to enable it on the NICs

2. If you plug a NIC to ANY hub you will not have "Full Duplex" mode (send and receive data simultaneously), even if the hub allows a speed of 100 Mbps (is 100BaseTx capable); this mode is only available for devices connected through a switch. SO:
2.1 plug all your servers in a switch if you want full duplex for the servers
2.2 if you plug workstations into J3295A machines, they will only run at "Half Duplex" (send data and only then receive data); plug any heavy duty workstation (yours, for instance) into the switch.
2.3 there is no point in having network printers (receive-only devices)on a switch - they can do well on hubs

[Ron Kinner]

Since your Win2K box says the network adaptor has been disconnected from the network and the network adaptor is connected to the hub and not the switch there should be no need to worry about the switch. Instead we need to worry about the direct link between the Win2K box and the hub.

This link has to be either 10 half or 100 half since that is all that the hub supports. If the NIC is set to 10 Full or 100 Full it will cause all sorts of problems so the first thing would be to verify what speed the NIC is running.

Generally you can do this in Win2K by Start/Settings/Network and Dialup Connections/Local Area Connections/Properties/

It should have a line about: Connect Using ..adaptor.

Make sure it is showing the one going to the hub if you have more than one. Now hit Configure. Usually with a modern NIC (or at least with one of the newer drivers) you will see either an indication of the speed and duplex on the General Page or a Details button which will show you what it is running at. If you don't see anything then you can at least check to see what it is set to do by going to Advanced. Usually there will be a list of features and you highlight the one you want and it will show what it is set at and allow you to set it differently. Verify that it is not set at FULL duplex.

Now Select the Driver tab and wait until it comes up withe the driver info. Note the maker, name and version number. For a built in Network Adaptor go to the PC maker's site and look for a new driver. For a separate card go to the card maker's site and see if there is a new driver. Install the new driver and see if that helps.

Also replace or test the cable, verify that the hub power is not going on and off and replace the hub with a different one or if you have the management interface on your hub you can check its status and log to see if you see anything.

What OS do the Vectras run?

Ron

[noka]

Hi Ron,
Thanks for reply but slight misunderstanding. There is no hub anymore, just ProCurve 2324 (J4818A) switches. PC's are set to auto (or 'hardware default' - some setting in NIC properties that I'm not sure about but I suspect it is same as 'auto'). Vectra VL400's and LC2000 run Win2000 SP2.

Rgds,
Norm

[Ron Kinner]

You said in your second post that the VL400's were connected to hubs. Since they exhibit the same symptoms it seems unlikely that the switch is to blame. However, it does not appear that the hubs are at fault either since you have other PCs connected directly to the switch.

SP3 has been out for quite a while, is quite stable and it fixes all sorts of nasties so it is time to bite the bullet and put it on everywhere. That appears to be one thing which is common to all of your boxes with problems so it is high on the list of probabilities.

Also 3Com has a really neat driver for their NICs. It give you all sorts of information and diagnostics and may fix a few problems. If you don't have their latest driver get it. (If the network interface is built in you have to get it from the PC maker). When you select the adapter and click on Configure it should come up with a display which shows you what duplex and speed you are running.

Of course it wouldn't hurt to upgrade your switch software as others have recommended but I don't think it will solve your current problem.

Ron

[noka]

Upgrading switch firmware for the three J4818A (ProCurve 2324) to F.05.14 appeared to stabilize the system.

Q: The three switches had older (and varying) firmware rev's. Must firmware of all switches be matched or different rev's be tolerated?

[Arimo Laine_2]

Hi again.

Differences in firmware revisions are tolerated, but for other reasons it's advisable to update them to the latest versions. Software download address is www.hp.com/rnd/software/switches.htm.

As for problems with console connection... dump Hyperterminal and go for TTPro instead, freeware and works with default settings with all HP Networking devices

Direct DL address http://hp.vector.co.jp/authors/VA002416/ttermp23.zip, homepage http://hp.vector.co.jp/authors/VA002416/teraterm.html.

Has extensions for SSH and SSL as well - and no, you don't have to install that japanese font :-)

Arimo

[Arimo Laine_2]

ARGH!

Darn forum application...

Download:
http://hp.vector.co.jp/authors/VA002416/ttermp23.zip

Homepage:
http://hp.vector.co.jp/authors/VA002416/teraterm.html

Arimo