visa open on gpib device takes 30 seconds after > 30 minute idle

[hbaugh]

After access to the device has been idle for > 30 minutes, it takes 30 seconds for visa open on the device to complete.

HW setup:

gpib device (either hp6674 power supply or hp function generator, doesn't matter)

GPIB-ENET100 ni488-2

linux rh3

ni visa 4.0 or ni visa 3.1

labview 7.1

This becomes a problem when long term tests are running and the timing of a particular test changes from run to run.

Is the visa rescanning for all devices?

Are there some properties that can be set to keep the device alive or active?

I used a vi that timed the open, write, read, close to the device and then waited a progressively longer period of time.  With repeats of < 30 minutes, the cycle would complete in 70-90 ms.  When repeats hit the 30 minute period, the cycle would complete in 30 seconds where the open took 30 seconds and the write/read/close took a negligible amount of time.

Hugh

[pBerg]

Howdy hbaugh,

You want to use a VISA Close at the end of each iteration of VISA communication. Then open a new VISA session using VISA Open for each test you run.

Do you see the same behavior when you run the test on another machine, perhaps Windows-based?

[hbaugh]

The test I am using is visa open/visa write/visa read/visa close wait 30 minutes, repeat.

When the wait is less than 30 minutes, the cycle take 50-70ms, greater than 30 minutes, the cycles take 30 seconds.

I have tried this on two different linux machines with similar setups with the same results.

I do not have access to a windows machine to try it out.

I was speculating it might be a gpib enet 100 thing.

Hugh

[pBerg]

Howdy hbaugh,

It sounds like your OS might have a TCP/IP timeout to the device if it happens at intervals >30min. I don't know how to change or check the timeout in RH, but it wouldn't surprise me if that were the culprit.

If it only happens if you wait >30min to read from the ENET device again, the simplest way to prevent a timeout would be to create an additional task that does some GPIB-ENET communication with the hardware at 15min after each test. You could do something simple like an *IDN? query that should keep the connection active and open without affecting anything your GPIB instruments are doing. Summarily, your timeline would look like this:

1. Perform your current test
2. Wait 15 min
3. Perform an *IDN? query
4. Wait 15 min.
5. Goto step 1 If not finished. Else, Goto step 6
6. End.

Using this technique the interval between communications to the GPIB-ENET would never exceed 30 min.

[hbaugh]

More info:

After > 30 minutes idle, even a ping to the gpib-enet100 box takes 20 seconds to respond which leads me to believe that the gpib-enet100 box itself it going into some sort of sleep mode.

I sent 4 consecutive pings (with default 10 second timeouts) and used tcpdump to observe all net traffic on that interface and saw the following:

Note: tennet is the interface that the gpib-enet100 box is on, i.e. 10.10.10.x

<first ping, times out>

10:38:43.121123 tennet > GPIBENET0: icmp: echo request (DF)

10:38:48.120396 arp who-has GPIBENET0 tell tennet

10:38:49.120379 arp who-has GPIBENET0 tell tennet

10:38:50.120373 arp who-has GPIBENET0 tell tennet

10:38:53.123511 arp who-has GPIBENET0 tell tennet

10:38:53.124043 arp reply GPIBENET0 is-at xx:xx:xx:xx:xx:xx

<second ping, times out>

10:38:53.124051 tennet > GPIBENET0: icmp: echo request (DF)

< third ping, receives response>

10:39:03.123446 tennet > GPIBENET0: icmp: echo request (DF)

10:39:03.124055 GPIBENET0 > tennet: icmp: echo reply

<fourth ping, receives response>

10:39:03.127233 tennet > GPIBENET0: icmp: echo request (DF)

10:39:03.127801 GPIBENET0 > tennet: icmp: echo reply

These results are repeatable after the gpib-enet100 box remains idle for > 30 minutes

My solution is to ping the gpib-enet100 box every 10 minutes to keep it awake which worked in overnight testing.

Did not see any settings that refer to a sleep mode but may be hidden.

This issue occurrs with both b9 and c9 versions of the firmware.

Any NI people out there that can comment on the gpib-enet100 going into a sleep mode after 30 minutes?

Hugh