Sixth test release of Tigon3 driver
0.94 turned out to be the "not quite Ginseng" release, but lets try to rectify that :-)
This is the "praying for powder snow up in Tahoe" release, or TAHOE for short. Get it at:
ftp://ftp.kernel.org/pub/linux/kernel/people/davem/TIGON3/tg3-0.95.patch.gz
[FIX] If we have to check the MAC_STAT register for link changes,
do not do it from interrupts. Do it from a timer instead. Hey maybe the Dell onboard Tigon3 problems are fixed for real now. :-)
[FIX] Off by one error in tg3_alloc_rx_skb() can lead to OOPS on
ifdown.
[FIX] 5700 chip interrupt status reporting protocol is racey,
work around it.
[DEBUG] Add assertions to TX ring processing.
[NICETY] When link comes up, report negotiated flow control settings. [FEATURE] Adaptive RX coalescing.
[FEATURE] Improved DMA bursting settings, major performance
improvement especially on non-x86 platforms (in particular Alpha and Sparc64 should get 10MB/sec extra bandwidth out of this).
[DOC] Add Configure.help entry for CONFIG_TIGON3
I'm really really really really really (NO, I MEAN REALLY) interested in performance comparisons of our 0.95 driver vs. Broadcom's stuff.
But, there is one special condition if you report benchmark numbers. If our driver is faster _AND_ you are an _Australian_ in _Australia_, you must send Jeff and I a case of Victoria Bitter for a job well done. :-))))))
Note that this thing is to Marcelo very soon. So get testing!
Franks a lot,
David S. Miller
[email protected]
David S. Miller:
Quoted 7 lines
I'm really really really really really (NO, I MEAN REALLY) interested in performance comparisons of our 0.95 driver vs. Broadcom's stuff. But, there is one special condition if you report benchmark numbers. If our driver is faster _AND_ you are an _Australian_ in _Australia_, you must send Jeff and I a case of Victoria Bitter for a job well done. :-))))))
Note that this thing is to Marcelo very soon. So get testing!
Still won't work with jumboframes and tcp.
Signature · 1 line
Thomas
From: Thomas Langås <[email protected]>
Date: Tue, 5 Mar 2002 14:35:19 +0100
Still won't work with jumboframes and tcp.
Which card do you have?
Also, can you try both changing the MTU during the initial up of the interface and later after the interface is up already? Thanks.
David S. Miller:
Quoted 4 lines
Wait, even more importantly are you going through a switch? Most gigabit switches don't support 9000 byte mtu :-) You're probably said that you're over a cross-over cable direct connect and I've just forgotten.
I'm going through a switch yes; a Cisco Catalyst 4006. I can have a chat with the networking folks here, but shouldn't the switch fragment the packet if it's too big?
Signature · 1 line
Thomas
From: Thomas Langås <[email protected]>
Date: Tue, 5 Mar 2002 15:19:36 +0100
I'm going through a switch yes; a Cisco Catalyst 4006. I can have a chat with the networking folks here, but shouldn't the switch fragment the packet if it's too big?
Switches then would need to understand IP, IPX, Netbeui... :-)
I hear some have tried it for at least IP, and admittedly my impression of the situation is a bit dated :)
David S. Miller:
Quoted 1 line
Most gigabit switches don't support 9000 byte mtu :-)
Hmm, I found a document through google; Cisco Catalyst 4006 doesn't support 9KB MTUs, so I'll contact the networking guys and fix this, we want switches that supports large MTUs :)
Signature · 1 line
Thomas
From: Thomas Langås <[email protected]>
Date: Tue, 5 Mar 2002 15:30:25 +0100
Hmm, I found a document through google; Cisco Catalyst 4006 doesn't support 9KB MTUs, so I'll contact the networking guys and fix this, we want switches that supports large MTUs :)
It's not a reasonable request, it doesn't interoperate at all with 10/100 segments, see my other mail.
Quoted 6 lines · --On Tuesday, March 5, 2002 3:30 PM +0100 Thomas Langås <[email protected]>…
David S. Miller:
Quoted 1 line
Most gigabit switches don't support 9000 byte mtu :-)
Hmm, I found a document through google; Cisco Catalyst 4006 doesn't support 9KB MTUs, so I'll contact the networking guys and fix this, we want switches that supports large MTUs :)
Good luck; they're fairly rare. From what I can tell, Cisco only supports jumbo frames on the Catalyst 5000 and 6000 families. Extreme supports them on at least a few models. The cheapest jumbo frame switch that I could find is the Intel 480T, which is over $7,000.
After doing a bunch of benchmarking, for our hardware and workload, jumbo frames don't really seem to help performance much anyway, so I ended up getting a $1,300 Dell 5012 (10 10/100/1000, 2 GBIC) switch.
I've heard rumors that Asante will have a jumbo-capable switch in May or so. Since they seem to OEM the same hardware that Dell does (for 3x Dell's cost), I wouldn't be too suprised to see a $1,500 16-port gig switch from Dell in a few months.
Scott
Quoted 11 lines · Scott Laird
--On Tuesday, March 5, 2002 3:30 PM +0100 Thomas Langås
Quoted 6 lines · <[email protected]> wrote:
David S. Miller:
Quoted 1 line
Most gigabit switches don't support 9000 byte mtu :-)
Hmm, I found a document through google; Cisco Catalyst 4006 doesn't support 9KB MTUs, so I'll contact the networking guys and fix this, we want switches that supports large MTUs :)
Good luck; they're fairly rare.
just an add-on:
NortelNetworks http://www.nortelnetworks.com sells Alteon Switches which are Jumbo-capable. I'm sure the're also available elsewhere. Don't know about the price though...
_sh_
David S. Miller:
Quoted 3 lines
How are you setting the mtu, with:
/sbin/ifconfig${DEV} mtu 9000
or something like that? Hmmm...
ifconfig eth1 up mtu 9000
and
ifconfig -a eth1 ... mtu 9000
both have been tried, no luck...
Signature · 1 line
Thomas
David S. Miller:
Quoted 1 line
Which card do you have?
01:08.0 Ethernet controller: BROADCOM Corporation BCM5700 1000BaseTX (rev 12)
Subsystem: Dell Computer Corporation: Unknown device 00d1
Flags: bus master, 66Mhz, medium devsel, latency 32, IRQ 17 Memory at feb00000 (64-bit, non-prefetchable) [size=64K]
Capabilities: [40] PCI-X non-bridge device.
Capabilities: [48] Power Management version 2
Capabilities: [50] Vital Product Data
Capabilities: [58] Message Signalled Interrupts: 64bit+ Queue=0/3 Enable-
That one :)
Quoted 3 lines
Also, can you try both changing the MTU during the initial up of the interface and later after the interface is up already? Thanks.
Did that, none work.
Signature · 1 line
Thomas
From: Thomas Langås <[email protected]>
Date: Tue, 5 Mar 2002 15:02:04 +0100
Quoted 3 lines
Also, can you try both changing the MTU during the initial up of the interface and later after the interface is up already? Thanks.
Did that, none work.
How are you setting the mtu, with:
/sbin/ifconfig ${DEV} mtu 9000
or something like that? Hmmm...
From: "David S. Miller" <[email protected]>
Date: Tue, 05 Mar 2002 06:03:23 -0800 (PST)
How are you setting the mtu, with:
/sbin/ifconfig ${DEV} mtu 9000or something like that? Hmmm...
Wait, even more importantly are you going through a switch?
Most gigabit switches don't support 9000 byte mtu :-)
You're probably said that you're over a cross-over cable direct connect and I've just forgotten.
David S. Miller:
Quoted 2 lines
It's not a reasonable request, it doesn't interoperate at all with 10/100 segments, see my other mail.
But when a load of our servers are on the same segment, they would benefit from 1000 Gbps internally, right?
How much of an effect would changing the MTU from 1k5 til 9k do in a scenario like that?
Signature · 1 line
Thomas