xmit timeouts with new e1000 driver

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



We have been using V3.6.8.1 of the e1000 driver
with no problems. Loading it with over 800Mbits/sec
causes no problems.

Now that the new e1000 driver, V4.3.15-k1, is part
of the linux sources I wanted to upgrade to the newer
code. In our basically "stock" 2.4.18 kernel, I added the
pci_mwi patch and the new e1000 sources from the BK tree.

We have tests that can generate a certain network load on
the server. Server transmits data using sendfile and we measure
the outbound/xmit data rate.

With the new (V4.3.15-k1) driver "right out of the box"
I got tranmit timeouts with only 100Mbits/sec load after running
for a minute or so. I noticed that the defaults for Rx descriptors
and RxIntDelay changed, so I loaded the new e1000 driver with the "old"
defaults for these params: Tx/Rx descriptors = 256, Tx/RxIntDelay=64.
I then retried the 100Mbit/sec load and it ran for 5 minutes or so
with no problems. I then restarted the test with a 400Mbit/sec load
and then I got a transmit timeout at about 200-250Mbits/sec as the load
ramped up.

I can use the exact same system and kernel, just using "old" V3.6.8.1
driver and run a load of about 825Mbits/sec for hours. The system we use
is based on a TYAN 2510 with 2x1GZ PIIIs, 4GB memory, and 2 e1000 NICs
(only 1 e1000 was active during testing).

The only change I made to the "new" driver, which is also in the "old" driver,
is that "watchdog timer" routine is now executed every 0.5 seconds instead of 2
seconds. Some stat reporting apps don't like stat update delays of more than a
second when reading stats.

So, what changed to make this driver more or less unusable for us?
I wanted to ask before spending more time (which is quite precious ATM).

Below is some output from "dmesg" with "old" and "new" driver. (The "old" driver
also has mods to dump params - helped when we were doing benchmarks last year).

thanks,

larry

Old driver:

Intel(R) PRO/1000 Network Driver - version 3.6.8.1
Copyright (c) 1999-2001 Intel Corporation.


Intel(R) PRO/1000 Network Connection
eth2: MAC Livengood (82543GC) Copper
eth2: Using value of 256 TxDescriptors
eth2: Using value of 256 RxDescriptors
eth2: Receive Checksum Offload Enabled
eth2: Flow Control Enabled
eth2: Using TxIntDelay of 64
eth2: Using RxIntDelay of 64
eth2: Mem:0xfebc0000  IRQ:27  Speed:N/A  Duplex:N/A
eth2: NIC Link is Up 1000 Mbps Full Duplex


New driver (default params):

Intel(R) PRO/1000 Network Driver - version 4.3.15-k1
Copyright (c) 1999-2002 Intel Corporation.
eth2: Intel(R) PRO/1000 Network Connection
e1000: eth2 NIC Link is Up 1000 Mbps Full Duplex
NETDEV WATCHDOG: eth2: transmit timed out
e1000: eth2 NIC Link is Up 1000 Mbps Full Duplex


New driver (params set to old defaults):

Intel(R) PRO/1000 Network Driver - version 4.3.15-k1
Copyright (c) 1999-2002 Intel Corporation.
eth2: Intel(R) PRO/1000 Network Connection
Transmit Descriptors set to 256
Receive Descriptors set to 256
Transmit Interrupt Delay set to 64
Transmit Absolute Interrupt Delay set to 64
Receive Interrupt Delay set to 64
Receive Absolute Interrupt Delay set to 64
NETDEV WATCHDOG: eth2: transmit timed out
e1000: eth2 NIC Link is Up 1000 Mbps Full Duplex


lspci -v output for the 2 e1000s in system:

01:02.0 Ethernet controller: Intel Corporation: Unknown device 1004 (rev 02)
        Subsystem: Intel Corporation: Unknown device 1004
        Flags: bus master, 66Mhz, medium devsel, latency 64, IRQ 27
        Memory at febc0000 (32-bit, non-prefetchable) [size=128K]
        Memory at febb0000 (32-bit, non-prefetchable) [size=64K]
        Expansion ROM at feba0000 [disabled] [size=64K]
        Capabilities: [dc] Power Management version 2

01:03.0 Ethernet controller: Intel Corporation: Unknown device 1004 (rev 02)
        Subsystem: Intel Corporation: Unknown device 1004
        Flags: bus master, 66Mhz, medium devsel, latency 64, IRQ 23
        Memory at feb80000 (32-bit, non-prefetchable) [size=128K]
        Memory at feb70000 (32-bit, non-prefetchable) [size=64K]
        Expansion ROM at feb60000 [disabled] [size=64K]
        Capabilities: [dc] Power Management version 2


-- 
--------------------------------------------------------------------------
  Larry Sendlosky                            larry.sendlosky@storigen.com
  Storigen Systems                                    (978) 323-4454 x261
--------------------------------------------------------------------------
-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[Index of Archives]     [Netdev]     [Ethernet Bridging]     [Linux 802.1Q VLAN]     [Linux Wireless]     [Kernel Newbies]     [Security]     [Linux for Hams]     [Netfilter]     [Git]     [Bugtraq]     [Yosemite News and Information]     [MIPS Linux]     [ARM Linux]     [Linux RAID]     [Linux PCI]     [Linux Admin]     [Samba]

  Powered by Linux