inconsistent server response

[Narayanan_2]

I have increased my network card speed from 10mbps to 100mbps full duplex. Ever since i am facing a slow/inconsistent server performance. I found this message in /var/adm/messages:

Feb 1 00:20:26 mysmpalpha vmunix: tu0: transmit FIFO underflow: threshold raised to: 256 bytes
Feb 3 00:23:31 mysmpalpha vmunix: tu0: transmit FIFO underflow: threshold raised to: 512 bytes

What does the message mean? How do i change network speed from hardware settings?

[Narayanan_2]

Forgot to add, the ping statistics:

64 bytes from 137.55.76.237: icmp_seq=200 ttl=128 time=10 ms
64 bytes from 137.55.76.237: icmp_seq=201 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=202 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=203 ttl=128 time=5 ms
64 bytes from 137.55.76.237: icmp_seq=204 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=205 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=206 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=207 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=208 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=209 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=210 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=211 ttl=128 time=6 ms
64 bytes from 137.55.76.237: icmp_seq=212 ttl=128 time=3 ms
64 bytes from 137.55.76.237: icmp_seq=213 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=214 ttl=128 time=8 ms
64 bytes from 137.55.76.237: icmp_seq=215 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=216 ttl=128 time=0 ms
64 bytes from 137.55.76.237: icmp_seq=217 ttl=128 time=1 ms

[Ralf Puchner]

this question was asked and answered several time within this forum. Please use the search page to find the answers.

the message means:
the adapter decided to switch to another DMA buffer size, leading to a reset of the adapter. Workaround is by setting a kernel parameter but this leads to more utilization for the CPU. Have a look into the "Tru64 network" forum for the details.

[Michael Schulte zur Sur]

Hi,

for more advice we need os release and patch kit and model of network card/switch. Ralf wants to encourage people to do a little research on their own before posting a question that has been answered on numerous occasions. But this would also mean less chances to earn points. ;-)
In ths case I would suggest to use the advanced search with the exact text "transmit FIFO underflow" and you will find a couple of threads, where this subject has been discussed and perhaps the solution for has been posted. ;-)
Anyway, come back, if you still have questions.

Michael

[Joris Denayer · Accepted solution]

Hi,

To anwer your 2nd question "How do i change network speed ...."

You should also verify the settings of the switch_port. Check if this port has fixed setting or works with auto-negotiation mode.


For the Tulip-based adapters (you have tu0), you can define the settings in SRM
>>> set ewa0_mode FastFD
A setting of "Auto-Negotiate" can also be used. A setting of "Auto-sense"
will not generally work with the DE500 series of Ethernet cards.
>>> set ewa0_mode auto

You can also use the lan_config command to modify the settings.

# lan_config
Usage: lan_config -i[interface] [-option]

Options:
-a[auto] Enable(1)/Disable(0) Auto Sense/Negotiation
-m[media] Media Type: (aui, bnc, utp, fiber)
-s[speed] Media Speed: (10, 20, 100, 200, 1000, 10000)
-x[duplex] Enable(1)/Disable(0) Full Duplex Mode


Try an ftp session to another 100Mbps connected system that is supposed to work correctly.
If the transfer rate is lower/equal then with 10Mbps settings, you have a mismatch half/full duplex.

Hope this helps

[Ralf Puchner]

1. ftp is not a performance messurement tool
2. if a speed mismatch occures, the netstat -i values (oerrs, colls, ierrs) dramatically increases and are the indication for a setup mismatch
3. within a twisted pair network all cables are "point-to-point" connections, so setup must be equal on both sides of the cable (autoneg/autoneg or fastfd/fastfd).

[Joris Denayer]

1) I agree
2) I agree, but not completely
3) This is indeed the answer for the second question of the original poster. He will be happy.