Mario Charest
06/11/2008 9:21 AM
post8973
|
Hello back again on the i82544 and Intel Pro 1000 PCIe.
Seems to work better when using transmit and receive descriptor at 1024 and the thread priority at 100.
However I still have some issue with what I think is the negotiation speed. And can't quite put my finger on it but
here is what I found so far; two computers connected to a cisco switch, there is nothing else on the switch.
Performance is bad, a pidin -n takes up to 5 seconds. nicinfo show not error and it report 1000m/bits full duplex so is
the switch. Then I try to plug the computers with a crossed cable, that doesn't work at all.
After rebooting the computers now I get the performance I'm expecting. I then connect the comptuers back to the switch
and now negotiation works ( those switch takes 30 to "negotiate"), but then the performance drops again.
Tried many switches/computers/network card ;-)
|
|
|
Hugh Brown
06/11/2008 9:34 AM
post8975
|
When I get some free time, I work on this driver to update it, but
unfortunately there is not much free time right now. If this is a
priority for you, then I suggest that you contact your sales rep to
escalate this problem.
-----Original Message-----
From: Mario Charest [mailto:mcharest@zinformatic.com]
Sent: Wednesday, June 11, 2008 9:22 AM
To: drivers-networking
Subject: i82544
Hello back again on the i82544 and Intel Pro 1000 PCIe.
Seems to work better when using transmit and receive descriptor at 1024
and the thread priority at 100.
However I still have some issue with what I think is the negotiation
speed. And can't quite put my finger on it but here is what I found so
far; two computers connected to a cisco switch, there is nothing else on
the switch. Performance is bad, a pidin -n takes up to 5 seconds.
nicinfo show not error and it report 1000m/bits full duplex so is the
switch. Then I try to plug the computers with a crossed cable, that
doesn't work at all.
After rebooting the computers now I get the performance I'm expecting.
I then connect the comptuers back to the switch and now negotiation
works ( those switch takes 30 to "negotiate"), but then the performance
drops again.
Tried many switches/computers/network card ;-)
_______________________________________________
Networking Drivers
http://community.qnx.com/sf/go/post8973
|
|
|
Mario Charest
06/11/2008 9:57 AM
post8979
|
> When I get some free time, I work on this driver to update it, but
> unfortunately there is not much free time right now. If this is a
> priority for you, then I suggest that you contact your sales rep to
> escalate this problem.
>
Thanks Hugh,
I was hoping that the negotiation part could be taken out of the context of a major rework of the driver ;)
Cheers,
Mario
|
|
|
Robert Craig
06/11/2008 10:11 AM
post8982
|
If you can try out io-pkt with the NetBSD wm driver and compare it to
the native devnp-i82544 driver, that might also give us some ideas as to
what to look for (although I'm not sure how right now :>)
Robert.
-----Original Message-----
From: Hugh Brown [mailto:hsbrown@qnx.com]
Sent: Wednesday, June 11, 2008 9:35 AM
To: drivers-networking
Subject: RE: i82544
When I get some free time, I work on this driver to update it, but
unfortunately there is not much free time right now. If this is a
priority for you, then I suggest that you contact your sales rep to
escalate this problem.
-----Original Message-----
From: Mario Charest [mailto:mcharest@zinformatic.com]
Sent: Wednesday, June 11, 2008 9:22 AM
To: drivers-networking
Subject: i82544
Hello back again on the i82544 and Intel Pro 1000 PCIe.
Seems to work better when using transmit and receive descriptor at 1024
and the thread priority at 100.
However I still have some issue with what I think is the negotiation
speed. And can't quite put my finger on it but here is what I found so
far; two computers connected to a cisco switch, there is nothing else on
the switch. Performance is bad, a pidin -n takes up to 5 seconds.
nicinfo show not error and it report 1000m/bits full duplex so is the
switch. Then I try to plug the computers with a crossed cable, that
doesn't work at all.
After rebooting the computers now I get the performance I'm expecting.
I then connect the comptuers back to the switch and now negotiation
works ( those switch takes 30 to "negotiate"), but then the performance
drops again.
Tried many switches/computers/network card ;-)
_______________________________________________
Networking Drivers
http://community.qnx.com/sf/go/post8973
_______________________________________________
Networking Drivers
http://community.qnx.com/sf/go/post8975
|
|
|
Mario Charest
06/11/2008 11:35 AM
post8992
|
> If you can try out io-pkt with the NetBSD wm driver and compare it to
> the native devnp-i82544 driver, that might also give us some ideas as to
> what to look for (although I'm not sure how right now :>
I'm on it.
>
> Robert.
>
> -----Original Message-----
> From: Hugh Brown [mailto:hsbrown@qnx.com]
> Sent: Wednesday, June 11, 2008 9:35 AM
> To: drivers-networking
> Subject: RE: i82544
>
> When I get some free time, I work on this driver to update it, but
> unfortunately there is not much free time right now. If this is a
> priority for you, then I suggest that you contact your sales rep to
> escalate this problem.
>
>
> -----Original Message-----
> From: Mario Charest [mailto:mcharest@zinformatic.com]
> Sent: Wednesday, June 11, 2008 9:22 AM
> To: drivers-networking
> Subject: i82544
>
>
> Hello back again on the i82544 and Intel Pro 1000 PCIe.
>
> Seems to work better when using transmit and receive descriptor at 1024
> and the thread priority at 100.
>
> However I still have some issue with what I think is the negotiation
> speed. And can't quite put my finger on it but here is what I found so
> far; two computers connected to a cisco switch, there is nothing else on
> the switch. Performance is bad, a pidin -n takes up to 5 seconds.
> nicinfo show not error and it report 1000m/bits full duplex so is the
> switch. Then I try to plug the computers with a crossed cable, that
> doesn't work at all.
>
> After rebooting the computers now I get the performance I'm expecting.
> I then connect the comptuers back to the switch and now negotiation
> works ( those switch takes 30 to "negotiate"), but then the performance
> drops again.
>
> Tried many switches/computers/network card ;-)
>
> _______________________________________________
> Networking Drivers
> http://community.qnx.com/sf/go/post8973
>
>
> _______________________________________________
> Networking Drivers
> http://community.qnx.com/sf/go/post8975
|
|
|
Mario Charest
06/11/2008 3:26 PM
post9024
|
> If you can try out io-pkt with the NetBSD wm driver and compare it to
> the native devnp-i82544 driver, that might also give us some ideas as to
> what to look for (although I'm not sure how right now :>)
>
I tried with the wm driver. This one I could switch to cross cable then back to the switch or force the switch to
100Mbits and it works fine. As for the performance it looks ok but within 2-3 minutes of light network activity ( pidin
-n ) io-pkt will sigsegv or it will display the "Out of interrupt event message"
I haven't tried devnp-i82544 yet.
- Mario
|
|
|
Mario Charest
06/11/2008 4:38 PM
post9028
|
> If you can try out io-pkt with the NetBSD wm driver and compare it to
> the native devnp-i82544 driver, that might also give us some ideas as to
> what to look for (although I'm not sure how right now :>)
I tried with both wm and devnp-i82544, both detect speed/duplex setting just fine.
devnp-i82544 is slow, even at 100Mbits ( so it's not timing issues with 1000Mbits )
with wm it seems fast as expected but hard to say because it crashes before I have time to run enough test.
I will try 6.3.2 devn-i82544 which is a pain because we have to specify the MAC addresses and device/vendor id, but
maybe it will give you guys some clues where to look.
Meanwhile I'll talk to our management to get them to talk with QSS sales rep to try to escalate this ( even send you
hardware )
|
|
|
Robert Craig
06/11/2008 4:45 PM
post9030
|
Try adding and irq_thresh=8000 as an option to the devnp-i82544 driver
and see if that changes the behaviour. This is the maximum number of
interrupts per second that the card will generate and it can make quite
a change to the performance characteristics. I'm surprised that the wm
driver is crashing. Is there anything from sloginfo that might point to
a problem?
Robert.
-----Original Message-----
From: Mario Charest [mailto:mcharest@zinformatic.com]
Sent: Wednesday, June 11, 2008 4:39 PM
To: drivers-networking
Subject: Re: RE: i82544
> If you can try out io-pkt with the NetBSD wm driver and compare it to
> the native devnp-i82544 driver, that might also give us some ideas as
to
> what to look for (although I'm not sure how right now :>)
I tried with both wm and devnp-i82544, both detect speed/duplex setting
just fine.
devnp-i82544 is slow, even at 100Mbits ( so it's not timing issues with
1000Mbits )
with wm it seems fast as expected but hard to say because it crashes
before I have time to run enough test.
I will try 6.3.2 devn-i82544 which is a pain because we have to specify
the MAC addresses and device/vendor id, but maybe it will give you guys
some clues where to look.
Meanwhile I'll talk to our management to get them to talk with QSS
sales rep to try to escalate this ( even send you hardware )
_______________________________________________
Networking Drivers
http://community.qnx.com/sf/go/post9028
|
|
|
Mario Charest
06/11/2008 5:23 PM
post9032
|
use devnp-wm.so doesn't describe any option. There must be some ?
> Try adding and irq_thresh=8000 as an option to the devnp-i82544 driver
> and see if that changes the behaviour. This is the maximum number of
> interrupts per second that the card will generate and it can make quite
> a change to the performance characteristics. I'm surprised that the wm
> driver is crashing. Is there anything from sloginfo that might point to
> a problem?
>
> Robert.
>
> -----Original Message-----
> From: Mario Charest [mailto:mcharest@zinformatic.com]
> Sent: Wednesday, June 11, 2008 4:39 PM
> To: drivers-networking
> Subject: Re: RE: i82544
>
> > If you can try out io-pkt with the NetBSD wm driver and compare it to
> > the native devnp-i82544 driver, that might also give us some ideas as
> to
> > what to look for (although I'm not sure how right now :>)
>
> I tried with both wm and devnp-i82544, both detect speed/duplex setting
> just fine.
>
> devnp-i82544 is slow, even at 100Mbits ( so it's not timing issues with
> 1000Mbits )
>
> with wm it seems fast as expected but hard to say because it crashes
> before I have time to run enough test.
>
> I will try 6.3.2 devn-i82544 which is a pain because we have to specify
> the MAC addresses and device/vendor id, but maybe it will give you guys
> some clues where to look.
>
> Meanwhile I'll talk to our management to get them to talk with QSS
> sales rep to try to escalate this ( even send you hardware )
>
> _______________________________________________
> Networking Drivers
> http://community.qnx.com/sf/go/post9028
|
|
|
Robert Craig
06/11/2008 5:35 PM
post9034
|
Right now, the ported netbsd drivers have no options. In the BSD world,
you can't pass in command line options since the drivers come up when
the kernel does. We're in the process of updating some of the
infrastructure to support some options generically and fully ported
drivers will have other options available.
Robert.
-----Original Message-----
From: Mario Charest [mailto:mcharest@zinformatic.com]
Sent: Wednesday, June 11, 2008 5:23 PM
To: drivers-networking
Subject: Re: RE: RE: i82544
use devnp-wm.so doesn't describe any option. There must be some ?
> Try adding and irq_thresh=8000 as an option to the devnp-i82544 driver
> and see if that changes the behaviour. This is the maximum number of
> interrupts per second that the card will generate and it can make
quite
> a change to the performance characteristics. I'm surprised that the
wm
> driver is crashing. Is there anything from sloginfo that might point
to
> a problem?
>
> Robert.
>
> -----Original Message-----
> From: Mario Charest [mailto:mcharest@zinformatic.com]
> Sent: Wednesday, June 11, 2008 4:39 PM
> To: drivers-networking
> Subject: Re: RE: i82544
>
> > If you can try out io-pkt with the NetBSD wm driver and compare it
to
> > the native devnp-i82544 driver, that might also give us some ideas
as
> to
> > what to look for (although I'm not sure how right now :>)
>
> I tried with both wm and devnp-i82544, both detect speed/duplex
setting
> just fine.
>
> devnp-i82544 is slow, even at 100Mbits ( so it's not timing issues
with
> 1000Mbits )
>
> with wm it seems fast as expected but hard to say because it crashes
> before I have time to run enough test.
>
> I will try 6.3.2 devn-i82544 which is a pain because we have to
specify
> the MAC addresses and device/vendor id, but maybe it will give you
guys
> some clues where to look.
>
> Meanwhile I'll talk to our management to get them to talk with QSS
> sales rep to try to escalate this ( even send you hardware )
>
> _______________________________________________
> Networking Drivers
> http://community.qnx.com/sf/go/post9028
_______________________________________________
Networking Drivers
http://community.qnx.com/sf/go/post9032
|
|
|
Mario Charest
06/11/2008 5:50 PM
post9035
|
> I'm surprised that the wm
> driver is crashing. Is there anything from sloginfo that might point to
> a problem?
Check the attached file:
>
> Robert.
>
|
|
|
Robert Craig
06/11/2008 5:56 PM
post9036
|
I see a culprit:
Jun 11 21:31:14 7 15 0 qnet(L4): l4_rx():
qos_verify_rx_conn_seq() failed: nd 37 conn 2
Jun 11 21:31:16 2 14 0 WARNING: mclpool limit reached;
increase NMBCLUSTERS
Jun 11 21:31:16 7 15 0 qnet(L4): en_iopkt_alloc_tx_pkt():
can't allocate mcluster for ifp 80deb40 (Not enough memory)
Now all we need to do is figure out why....
Robert.
-----Original Message-----
From: Mario Charest [mailto:mcharest@zinformatic.com]
Sent: Wednesday, June 11, 2008 5:51 PM
To: drivers-networking
Subject: Re: RE: RE: i82544
> I'm surprised that the wm
> driver is crashing. Is there anything from sloginfo that might point
to
> a problem?
Check the attached file:
>
> Robert.
>
_______________________________________________
Networking Drivers
http://community.qnx.com/sf/go/post9035
|
|
|
Mario Charest
06/12/2008 9:44 AM
post9044
|
>
I found a work around. We also use a dual port version of that card ( VID 8086, DID = 105e ). Out of desperation I
tried it out and so far everything looks good ( with devn-i82544).
If you would like me to run some test for the devnp-wm and the single port card I'd be happy to contribute ;-)
javascript:if%20(validateCreatePost(document.createPost))%20{%20submitForm(document.createPost,%20'submit');%20}%20else%
20{%20void(0)%20}
> I see a culprit:
>
> Jun 11 21:31:14 7 15 0 qnet(L4): l4_rx():
> qos_verify_rx_conn_seq() failed: nd 37 conn 2
>
> Jun 11 21:31:16 2 14 0 WARNING: mclpool limit reached;
> increase NMBCLUSTERS
>
> Jun 11 21:31:16 7 15 0 qnet(L4): en_iopkt_alloc_tx_pkt():
> can't allocate mcluster for ifp 80deb40 (Not enough memory)
>
>
> Now all we need to do is figure out why....
>
>
> Robert.
>
> -----Original Message-----
> From: Mario Charest [mailto:mcharest@zinformatic.com]
> Sent: Wednesday, June 11, 2008 5:51 PM
> To: drivers-networking
> Subject: Re: RE: RE: i82544
>
> > I'm surprised that the wm
> > driver is crashing. Is there anything from sloginfo that might point
> to
> > a problem?
>
> Check the attached file:
>
> >
> > Robert.
> >
>
> _______________________________________________
> Networking Drivers
> http://community.qnx.com/sf/go/post9035
|
|
|
Hugh Brown
06/12/2008 3:41 PM
post9065
|
Hi Mario,
Please will you give the attached early experimental driver a test and
let me know how it works. I have changed the name to devn-e1000.so, as
this is a port of the Intel driver. This driver will also work with the
i82544.
Thanks, Hugh.
-----Original Message-----
From: Mario Charest [mailto:mcharest@zinformatic.com]
Sent: Thursday, June 12, 2008 9:44 AM
To: drivers-networking
Subject: Re: RE: RE: RE: i82544
>
I found a work around. We also use a dual port version of that card (
VID 8086, DID = 105e ). Out of desperation I tried it out and so far
everything looks good ( with devn-i82544).
If you would like me to run some test for the devnp-wm and the single
port card I'd be happy to contribute ;-)
javascript:if%20(validateCreatePost(document.createPost))%20{%20submitFo
rm(document.createPost,%20'submit');%20}%20else%20{%20void(0)%20}
> I see a culprit:
>
> Jun 11 21:31:14 7 15 0 qnet(L4): l4_rx():
> qos_verify_rx_conn_seq() failed: nd 37 conn 2
>
> Jun 11 21:31:16 2 14 0 WARNING: mclpool limit reached;
> increase NMBCLUSTERS
>
> Jun 11 21:31:16 7 15 0 qnet(L4): en_iopkt_alloc_tx_pkt():
> can't allocate mcluster for ifp 80deb40 (Not enough memory)
>
>
> Now all we need to do is figure out why....
>
>
> Robert.
>
> -----Original Message-----
> From: Mario Charest [mailto:mcharest@zinformatic.com]
> Sent: Wednesday, June 11, 2008 5:51 PM
> To: drivers-networking
> Subject: Re: RE: RE: i82544
>
> > I'm surprised that the wm
> > driver is crashing. Is there anything from sloginfo that might point
> to
> > a problem?
>
> Check the attached file:
>
> >
> > Robert.
> >
>
> _______________________________________________
> Networking Drivers
> http://community.qnx.com/sf/go/post9035
_______________________________________________
Networking Drivers
http://community.qnx.com/sf/go/post9044
|
|
|
Robert Craig
06/12/2008 10:34 AM
post9049
|
Hi Mario:
Can you try out the attached io-pkt driver (x86) and see if the
performance is better? Don't forget the irq_thresh=8000 option.
Robert.
-----Original Message-----
From: Mario Charest [mailto:mcharest@zinformatic.com]
Sent: Wednesday, June 11, 2008 4:39 PM
To: drivers-networking
Subject: Re: RE: i82544
> If you can try out io-pkt with the NetBSD wm driver and compare it to
> the native devnp-i82544 driver, that might also give us some ideas as
to
> what to look for (although I'm not sure how right now :>)
I tried with both wm and devnp-i82544, both detect speed/duplex setting
just fine.
devnp-i82544 is slow, even at 100Mbits ( so it's not timing issues with
1000Mbits )
with wm it seems fast as expected but hard to say because it crashes
before I have time to run enough test.
I will try 6.3.2 devn-i82544 which is a pain because we have to specify
the MAC addresses and device/vendor id, but maybe it will give you guys
some clues where to look.
Meanwhile I'll talk to our management to get them to talk with QSS
sales rep to try to escalate this ( even send you hardware )
_______________________________________________
Networking Drivers
http://community.qnx.com/sf/go/post9028
|
|
|
|