Thread: [mpls-linux-devel] Kernel 2.6.32.11 or better
Status: Beta
Brought to you by:
jleu
Menu
▾
▴
mpls-linux-devel
|
From: Chris R. <Chr...@nr...> - 2010-04-19 13:29:53
|
Hi Jim I was wondering if you are working to get mpls-linux updated to at least kernel 2.6.32-11 (aka the latest Fedora 12 update)? I have several Dell 610 Nehalem CPU based system to test MPLS-Linux 10GE performance on but all are running FC12. Non-MPLS path 10GE performance improvements on Nehalem systems is remarkable. Typical Quad-Core systems get about 5G or 6G but with the Nehalem I'm seeing almost line rate at about 9.8G sustained for non-MPLS test throughput. Thus I'd like to see how the MPLS-Linux can improve over past performance test. I have been working a new patch but as in the past there are dramatic changes to the kernel from 2.6.27 (current MPLS git level) to 2.6.32-11. Notably with skbuff, which so far my changes are failing miserably. ...Chris |
|
From: <jl...@mi...> - 2010-04-21 04:32:45
|
I will take a look at Renato's patch and go from there. On Mon, Apr 19, 2010 at 09:06:05AM -0400, Chris Robson wrote: > > Hi Jim > > I was wondering if you are working to get mpls-linux updated to at least > kernel 2.6.32-11 (aka the latest Fedora 12 update)? > > I have several Dell 610 Nehalem CPU based system to test MPLS-Linux 10GE > performance on but all are running FC12. > > Non-MPLS path 10GE performance improvements on Nehalem systems is > remarkable. Typical Quad-Core systems get about 5G or 6G but with the > Nehalem I'm seeing almost line rate at about 9.8G sustained for non-MPLS > test throughput. Thus I'd like to see how the MPLS-Linux can improve > over past performance test. > > I have been working a new patch but as in the past there are dramatic > changes to the kernel from 2.6.27 (current MPLS git level) to > 2.6.32-11. Notably with skbuff, which so far my changes are failing > miserably. > > ...Chris > > ------------------------------------------------------------------------------ > Download Intel® Parallel Studio Eval > Try the new software tools for yourself. Speed compiling, find bugs > proactively, and fine-tune applications for parallel performance. > See why Intel Parallel Studio got high marks during beta. > http://p.sf.net/sfu/intel-sw-dev > _______________________________________________ > mpls-linux-devel mailing list > mpl...@li... > https://lists.sourceforge.net/lists/listinfo/mpls-linux-devel -- James R. Leu jl...@mi... |
|
From: Renato W. <ren...@gm...> - 2010-04-19 16:07:47
|
Hi, I have already ported mpls-1.963 to the kernel 2.6.32.11, I just didn't ported the ebtables module. The only problem is that you should disable the stack protector on the kernel, since a buffer overflow occurs when you bind a route with a nhlfe key. I'm working on this bug, but disabling stack protector everything should work. Patch for kernel 2.6.32.11: 1 - Get kernel 2.6.32.11: http://www.kernel.org/pub/linux/kernel/v2.6/linux-2.6.32.11.tar.bz2 2 - Apply attached patch: patch -p1 < ../linux-2.6.32.11-mpls1.963.patch 3 - make menuconfig: Networking options ---> <*> Multiprotocol Label Switching <*> MPLS: Virtual tunnel interface <*> IP: MPLS support <*> The IPv6 protocol ---> <*> IPv6: MPLS support [*] Network packet filtering framework (Netfilter) ---> Core Netfilter Configuration ---> <*> "mpls" target support General setup ---> [*] Network namespace Disable stack protector Processor type and features ---> [ ] Enable -fstack-protector buffer overflow detection (EXPERIMENTAL) 4 - build kernel... Patch for iproute2: 1 - Get iproute2-2.6.33: http://devresources.linuxfoundation.org/dev/iproute2/download/iproute2-2.6.33.tar.bz2 2 - Apply attached patch: patch -p1 < ../iproute2-2.6.33-mpls1.963.patch 3 - $ make # make install (*) * build dependencies (debian5): libdb-dev, bison, flex Patch for iptables: 1 - Get iptables-1.4.7: http://www.netfilter.org/projects/iptables/files/iptables-1.4.7.tar.bz2 2 - Apply attached patch: patch -p1 < ../iptables-1.4.7-mpls1.963.patch 3 - $ ./configure $ make # make install Greetings from Brazil, Renato. . .2010/4/19 Chris Robson <Chr...@nr...> > > Hi Jim > > I was wondering if you are working to get mpls-linux updated to at least > kernel 2.6.32-11 (aka the latest Fedora 12 update)? > > I have several Dell 610 Nehalem CPU based system to test MPLS-Linux 10GE > performance on but all are running FC12. > > Non-MPLS path 10GE performance improvements on Nehalem systems is > remarkable. Typical Quad-Core systems get about 5G or 6G but with the > Nehalem I'm seeing almost line rate at about 9.8G sustained for non-MPLS > test throughput. Thus I'd like to see how the MPLS-Linux can improve > over past performance test. > > I have been working a new patch but as in the past there are dramatic > changes to the kernel from 2.6.27 (current MPLS git level) to > 2.6.32-11. Notably with skbuff, which so far my changes are failing > miserably. > > ...Chris > > > ------------------------------------------------------------------------------ > Download Intel® Parallel Studio Eval > Try the new software tools for yourself. Speed compiling, find bugs > proactively, and fine-tune applications for parallel performance. > See why Intel Parallel Studio got high marks during beta. > http://p.sf.net/sfu/intel-sw-dev > _______________________________________________ > mpls-linux-devel mailing list > mpl...@li... > https://lists.sourceforge.net/lists/listinfo/mpls-linux-devel > |
|
From: Chris R. <Chr...@nr...> - 2010-05-04 10:46:14
|
Renato Based on a quick check, the patches I came up with are almost (about 75%) the same as yours. But since they didnt match completely I attempted to build a new FC12 rpm using your kernel and iproute patches. Unfortunately the system locks up on receipt of the 1st MPLS packet just like a patched kernel using my patches. I check if the build disables the stack overflow and it turns out the flag CONFIG_CC_STACKPROTECTOR in the file config-generic is not even present which I "think" means it's not included in the compile by default. So there is something else wrong and as I said before, since the major changes from 2.6.27 and 2.6.32 appears to be with the skbuff I suspect something there is the problem. Also, there seems to be some interesting changes with MPLS inclusion into the FC12 rpm build files. I had to remove patches that add the ipv4/Makefile changes because the mpls.o file inclusion is already there. This is just one example of some additions to the make scripts which include MPLS but note the actual mpls code is not present, ergo a bit confusing as to what is going on? ....Chris On 04/19/2010 12:07 PM, Renato Westphal wrote: > Hi, > > I have already ported mpls-1.963 to the kernel 2.6.32.11, I just > didn't ported the ebtables module. > > The only problem is that you should disable the stack protector on the > kernel, > since a buffer overflow occurs when you bind a route with a nhlfe key. > I'm working on this bug, but disabling stack protector everything > should work. > > Patch for kernel 2.6.32.11 <http://2.6.32.11>: > > 1 - Get kernel 2.6.32.11 <http://2.6.32.11>: > http://www.kernel.org/pub/linux/kernel/v2.6/linux-2.6.32.11.tar.bz2 > > 2 - Apply attached patch: patch -p1 < ../linux-2.6.32.11-mpls1.963.patch > > 3 - make menuconfig: > > Networking options ---> > <*> Multiprotocol Label Switching > <*> MPLS: Virtual tunnel interface > <*> IP: MPLS support > <*> The IPv6 protocol ---> > <*> IPv6: MPLS support > > [*] Network packet filtering framework (Netfilter) ---> > Core Netfilter Configuration ---> > <*> "mpls" target support > > General setup ---> > [*] Network namespace > > Disable stack protector > > Processor type and features ---> > [ ] Enable -fstack-protector buffer overflow detection (EXPERIMENTAL) > > 4 - build kernel... > > > Patch for iproute2: > > 1 - Get iproute2-2.6.33: > http://devresources.linuxfoundation.org/dev/iproute2/download/iproute2-2.6.33.tar.bz2 > > 2 - Apply attached patch: patch -p1 < ../iproute2-2.6.33-mpls1.963.patch > > 3 - $ make > # make install (*) > > * build dependencies (debian5): libdb-dev, bison, flex > > > Patch for iptables: > > 1 - Get iptables-1.4.7: > http://www.netfilter.org/projects/iptables/files/iptables-1.4.7.tar.bz2 > > 2 - Apply attached patch: patch -p1 < ../iptables-1.4.7-mpls1.963.patch > > 3 - $ ./configure > $ make > # make install > > > Greetings from Brazil, > Renato. > > . > .2010/4/19 Chris Robson <Chr...@nr... > <mailto:Chr...@nr...>> > > > Hi Jim > > I was wondering if you are working to get mpls-linux updated to at > least > kernel 2.6.32-11 (aka the latest Fedora 12 update)? > > I have several Dell 610 Nehalem CPU based system to test > MPLS-Linux 10GE > performance on but all are running FC12. > > Non-MPLS path 10GE performance improvements on Nehalem systems is > remarkable. Typical Quad-Core systems get about 5G or 6G but with the > Nehalem I'm seeing almost line rate at about 9.8G sustained for > non-MPLS > test throughput. Thus I'd like to see how the MPLS-Linux can improve > over past performance test. > > I have been working a new patch but as in the past there are dramatic > changes to the kernel from 2.6.27 (current MPLS git level) to > 2.6.32-11. Notably with skbuff, which so far my changes are failing > miserably. > > ...Chris > > ------------------------------------------------------------------------------ > Download Intel® Parallel Studio Eval > Try the new software tools for yourself. Speed compiling, find bugs > proactively, and fine-tune applications for parallel performance. > See why Intel Parallel Studio got high marks during beta. > http://p.sf.net/sfu/intel-sw-dev > _______________________________________________ > mpls-linux-devel mailing list > mpl...@li... > <mailto:mpl...@li...> > https://lists.sourceforge.net/lists/listinfo/mpls-linux-devel > > |
|
From: Chris R. <Chr...@nr...> - 2010-05-05 11:41:13
|
Jorge
Testing today shows that the lock up problem still exist after applying
the patch you sent out. Unfortunately I can't give you a tcpdump from
Host A cause the system locks up tight and requires a hard power reset.
The OS I'm using is FC12 version 2.6.32.11-99.fc12.x86_64 modified with
the Renato patch and your additional patch. If you want some additional
info I can re-run test but please provide specifics to how/what/where
you need things done, remembering I'm not an expert in system level
debugging but can parakeet command requires fairly good (:-) *sigh*.
Maybe I show use a patch file created by you as a first step? Can you
send me something direct?
Data and config info:
Host A (eth1/10.50.1.1 - eth1:1/10.50.210.1)<---->Host B(eth1/10.50.1.2
- eth1:1/10.50.110.1)
Host A MPLS manual configuration:
/sbin/mpls labelspace set dev eth1 labelspace 0
LSPKey=`/sbin/mpls nhlfe add key 0 instructions push gen 1000 nexthop
eth1 ipv4 10.50.1.2 | grep key | cut -c 17-26`
/sbin/ip route add 10.50.110.0/24 via 10.50.1.2 mpls $LSPKey
/sbin/mpls ilm add label gen 2000 labelspace 0
ILM entry label gen 2000 labelspace 0 proto ipv4
pop peek (0 bytes, 0 pkts)
NHLFE entry key 0x00000002 mtu 1496 propagate_ttl
push gen 1000 set eth1 ipv4 10.50.1.2 (0 bytes, 0 pkts)
LABELSPACE entry dev lo labelspace -1
LABELSPACE entry dev mpls0 labelspace -1
LABELSPACE entry dev eth0 labelspace -1
LABELSPACE entry dev eth1 labelspace 0
Host A interface config info
eth1 Link encap:Ethernet HWaddr 00:05:1B:70:E9:37
inet addr:10.50.1.1 Bcast:10.50.1.255 Mask:255.255.255.0
inet6 addr: fe80::205:1bff:fe70:e937/64 Scope:Link
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:1 errors:0 dropped:0 overruns:0 frame:0
TX packets:48 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:437 (437.0 b) TX bytes:7040 (6.8 KiB)
eth1:1 Link encap:Ethernet HWaddr 00:05:1B:70:E9:37
inet addr:10.50.210.1 Bcast:10.50.210.255 Mask:255.255.255.0
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
Host A route info:
10.50.110.0/24 via 10.50.1.2 dev eth1 mpls 0x2
On 05/04/2010 08:10 AM, Jorge Boncompte [DTI2] wrote:
> El 04/05/2010 12:45, Chris Robson escribió:
>
>> Renato
>>
>> Based on a quick check, the patches I came up with are almost (about
>> 75%) the same as yours. But since they didnt match completely I
>> attempted to build a new FC12 rpm using your kernel and iproute
>> patches. Unfortunately the system locks up on receipt of the 1st MPLS
>> packet just like a patched kernel using my patches. I check if the
>> build disables the stack overflow and it turns out the flag
>> CONFIG_CC_STACKPROTECTOR in the file config-generic is not even present
>> which I "think" means it's not included in the compile by default. So
>> there is something else wrong and as I said before, since the major
>> changes from 2.6.27 and 2.6.32 appears to be with the skbuff I suspect
>> something there is the problem.
>>
>> Also, there seems to be some interesting changes with MPLS inclusion
>> into the FC12 rpm build files. I had to remove patches that add the
>> ipv4/Makefile changes because the mpls.o file inclusion is already
>> there. This is just one example of some additions to the make scripts
>> which include MPLS but note the actual mpls code is not present, ergo a
>> bit confusing as to what is going on?
>>
> Hi, could you test with the attached patch again? Or send me a decoded oops and
> some instructions to replicate your scenario!
>
>
|
|
From: Jorge B. [DTI2] <jo...@dt...> - 2010-05-05 13:24:39
|
El 05/05/2010 13:40, Chris Robson escribió: > > Jorge > > Testing today shows that the lock up problem still exist after applying the > patch you sent out. Unfortunately I can't give you a tcpdump from Host A > cause the system locks up tight and requires a hard power reset. The OS I'm > using is FC12 version 2.6.32.11-99.fc12.x86_64 modified with the Renato patch > and your additional patch. If you want some additional info I can re-run > test but please provide specifics to how/what/where you need things done, > remembering I'm not an expert in system level debugging but can parakeet > command requires fairly good (:-) *sigh*. Maybe I show use a patch file > created by you as a first step? Can you send me something direct? I am on Debian and I only use kernel.org kernels. I have tested in my 2.6.33.2 setup with your example and it does work. I haven't seen Renato's patch nor Pali patches you did mention in the other email. If you can send me a diff between your current tree and the original Fedora one maybe I can pinpoint the differences. Regards, Jorge -- ============================================================== Jorge Boncompte - Ingenieria y Gestion de RED DTI2 - Desarrollo de la Tecnologia de las Comunicaciones -------------------------------------------------------------- C/ Abogado Enriquez Barrios, 5 14004 CORDOBA (SPAIN) Tlf: +34 957 761395 / FAX: +34 957 450380 ============================================================== - Sin pistachos no hay Rock & Roll... - Without wicker a basket cannot be made. ============================================================== |
×
Want the latest updates on software, tech news, and AI?
Get latest updates about software, tech news, and AI from SourceForge directly in your inbox once a month.
Thanks for helping keep SourceForge clean.
X