[LTP] NUMA system and get_a_numa_node()

[LTP] NUMA system and get_a_numa_node()

[Han Pingtian]

Hey there,

We have a system which only has one NUMA node. When trying to run ksm02
on it, becase the code calling get_allowed_nodes() with a count 2 as
below:

 972 /* Warning: *DO NOT* use this function in child */
 973 unsigned int get_a_numa_node(void (*cleanup_fn) (void))
 974 {              
 975         unsigned int nd1, nd2;
 976         int ret; 
 977                        
 978         ret = get_allowed_nodes(0, 2, &nd1, &nd2);
 979         switch (ret) { 
 980         case 0:
 981                 break;
 982         case -3:
 983                 tst_brkm(TCONF, cleanup_fn, "requires a NUMA system.");

it always failed with "requires a NUMA system.".

I think a NUMA system may have only one node, right? If so, could we
change the count to 1?

Thanks in advance!


------------------------------------------------------------------------------
Comprehensive Server Monitoring with Site24x7.
Monitor 10 servers for $9/Month.
Get alerted through email, SMS, voice calls or mobile push notifications.
Take corrective actions from your mobile device.
http://pubads.g.doubleclick.net/gampad/clk?id=154624111&iu=/4140/ostg.clktrk
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] NUMA system and get_a_numa_node()

[Cyril Hrubis]

Hi!
> We have a system which only has one NUMA node. When trying to run ksm02
> on it, becase the code calling get_allowed_nodes() with a count 2 as
> below:
> 
>  972 /* Warning: *DO NOT* use this function in child */
>  973 unsigned int get_a_numa_node(void (*cleanup_fn) (void))
>  974 {              
>  975         unsigned int nd1, nd2;
>  976         int ret; 
>  977                        
>  978         ret = get_allowed_nodes(0, 2, &nd1, &nd2);
>  979         switch (ret) { 
>  980         case 0:
>  981                 break;
>  982         case -3:
>  983                 tst_brkm(TCONF, cleanup_fn, "requires a NUMA system.");
> 
> it always failed with "requires a NUMA system.".
> 
> I think a NUMA system may have only one node, right? If so, could we
> change the count to 1?

Hmm, are you sure that the testcases are actually applicable to a system
with only one node?

Because what the testcases do in the setup is to write 1 to
/sys/kernel/mm/ksm/merge_across_nodes so I would expect that the test
needs at least two nodes...

-- 
Cyril Hrubis
chrubis@suse.cz

------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=157005751&iu=/4140/ostg.clktrk
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] NUMA system and get_a_numa_node()

[Han Pingtian]

On Wed, Nov 26, 2014 at 01:53:36PM +0100, Cyril Hrubis wrote:
> Hi!
> > We have a system which only has one NUMA node. When trying to run ksm02
> > on it, becase the code calling get_allowed_nodes() with a count 2 as
> > below:
> > 
> >  972 /* Warning: *DO NOT* use this function in child */
> >  973 unsigned int get_a_numa_node(void (*cleanup_fn) (void))
> >  974 {              
> >  975         unsigned int nd1, nd2;
> >  976         int ret; 
> >  977                        
> >  978         ret = get_allowed_nodes(0, 2, &nd1, &nd2);
> >  979         switch (ret) { 
> >  980         case 0:
> >  981                 break;
> >  982         case -3:
> >  983                 tst_brkm(TCONF, cleanup_fn, "requires a NUMA system.");
> > 
> > it always failed with "requires a NUMA system.".
> > 
> > I think a NUMA system may have only one node, right? If so, could we
> > change the count to 1?
> 
> Hmm, are you sure that the testcases are actually applicable to a system
> with only one node?
> 
> Because what the testcases do in the setup is to write 1 to
> /sys/kernel/mm/ksm/merge_across_nodes so I would expect that the test
> needs at least two nodes...
> 
I think if there is only one NUMA node, then this is a special case for
this testcase, it'll test if set "merge_across_nodes" to 1, any problem
will be triggered.

I have tried to change the node count to 1 in get_allowed_nodes() and
recompiled ksm02. It works successfully.

Thanks.


------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=157005751&iu=/4140/ostg.clktrk
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] NUMA system and get_a_numa_node()

[Wanlong Gao]

On 11/27/2014 10:42 AM, Han Pingtian wrote:
> On Wed, Nov 26, 2014 at 01:53:36PM +0100, Cyril Hrubis wrote:
>> Hi!
>>> We have a system which only has one NUMA node. When trying to run ksm02
>>> on it, becase the code calling get_allowed_nodes() with a count 2 as
>>> below:
>>>
>>>  972 /* Warning: *DO NOT* use this function in child */
>>>  973 unsigned int get_a_numa_node(void (*cleanup_fn) (void))
>>>  974 {              
>>>  975         unsigned int nd1, nd2;
>>>  976         int ret; 
>>>  977                        
>>>  978         ret = get_allowed_nodes(0, 2, &nd1, &nd2);
>>>  979         switch (ret) { 
>>>  980         case 0:
>>>  981                 break;
>>>  982         case -3:
>>>  983                 tst_brkm(TCONF, cleanup_fn, "requires a NUMA system.");
>>>
>>> it always failed with "requires a NUMA system.".
>>>
>>> I think a NUMA system may have only one node, right? If so, could we
>>> change the count to 1?
>>
>> Hmm, are you sure that the testcases are actually applicable to a system
>> with only one node?
>>
>> Because what the testcases do in the setup is to write 1 to
>> /sys/kernel/mm/ksm/merge_across_nodes so I would expect that the test
>> needs at least two nodes...
>>
> I think if there is only one NUMA node, then this is a special case for
> this testcase, it'll test if set "merge_across_nodes" to 1, any problem
> will be triggered.
> 
> I have tried to change the node count to 1 in get_allowed_nodes() and
> recompiled ksm02. It works successfully.
> 


Do you think ONLY ONE NODE can be called NUMA?

http://en.wikipedia.org/wiki/Non-uniform_memory_access

Thanks,
Wanlong Gao

> Thanks.
> 
> 
> ------------------------------------------------------------------------------
> Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
> from Actuate! Instantly Supercharge Your Business Reports and Dashboards
> with Interactivity, Sharing, Native Excel Exports, App Integration & more
> Get technology previously reserved for billion-dollar corporations, FREE
> http://pubads.g.doubleclick.net/gampad/clk?id=157005751&iu=/4140/ostg.clktrk
> _______________________________________________
> Ltp-list mailing list
> Ltp-list@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/ltp-list
> 


------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=157005751&iu=/4140/ostg.clktrk
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] NUMA system and get_a_numa_node()

[Han Pingtian]

On Thu, Nov 27, 2014 at 11:06:15AM +0800, Wanlong Gao wrote:
> On 11/27/2014 10:42 AM, Han Pingtian wrote:
> > On Wed, Nov 26, 2014 at 01:53:36PM +0100, Cyril Hrubis wrote:
> >> Hi!
> >>> We have a system which only has one NUMA node. When trying to run ksm02
> >>> on it, becase the code calling get_allowed_nodes() with a count 2 as
> >>> below:
> >>>
> >>>  972 /* Warning: *DO NOT* use this function in child */
> >>>  973 unsigned int get_a_numa_node(void (*cleanup_fn) (void))
> >>>  974 {              
> >>>  975         unsigned int nd1, nd2;
> >>>  976         int ret; 
> >>>  977                        
> >>>  978         ret = get_allowed_nodes(0, 2, &nd1, &nd2);
> >>>  979         switch (ret) { 
> >>>  980         case 0:
> >>>  981                 break;
> >>>  982         case -3:
> >>>  983                 tst_brkm(TCONF, cleanup_fn, "requires a NUMA system.");
> >>>
> >>> it always failed with "requires a NUMA system.".
> >>>
> >>> I think a NUMA system may have only one node, right? If so, could we
> >>> change the count to 1?
> >>
> >> Hmm, are you sure that the testcases are actually applicable to a system
> >> with only one node?
> >>
> >> Because what the testcases do in the setup is to write 1 to
> >> /sys/kernel/mm/ksm/merge_across_nodes so I would expect that the test
> >> needs at least two nodes...
> >>
> > I think if there is only one NUMA node, then this is a special case for
> > this testcase, it'll test if set "merge_across_nodes" to 1, any problem
> > will be triggered.
> > 
> > I have tried to change the node count to 1 in get_allowed_nodes() and
> > recompiled ksm02. It works successfully.
> > 
> 
> 
> Do you think ONLY ONE NODE can be called NUMA?
> 
> http://en.wikipedia.org/wiki/Non-uniform_memory_access
> 

I think it's not a problem to run the testcase on a non-NUMA system, if
we take it as a special case?

Thanks.


------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=157005751&iu=/4140/ostg.clktrk
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] NUMA system and get_a_numa_node()

[Wanlong Gao]

On 11/27/2014 03:45 PM, Han Pingtian wrote:
> On Thu, Nov 27, 2014 at 11:06:15AM +0800, Wanlong Gao wrote:
>> On 11/27/2014 10:42 AM, Han Pingtian wrote:
>>> On Wed, Nov 26, 2014 at 01:53:36PM +0100, Cyril Hrubis wrote:
>>>> Hi!
>>>>> We have a system which only has one NUMA node. When trying to run ksm02
>>>>> on it, becase the code calling get_allowed_nodes() with a count 2 as
>>>>> below:
>>>>>
>>>>>  972 /* Warning: *DO NOT* use this function in child */
>>>>>  973 unsigned int get_a_numa_node(void (*cleanup_fn) (void))
>>>>>  974 {              
>>>>>  975         unsigned int nd1, nd2;
>>>>>  976         int ret; 
>>>>>  977                        
>>>>>  978         ret = get_allowed_nodes(0, 2, &nd1, &nd2);
>>>>>  979         switch (ret) { 
>>>>>  980         case 0:
>>>>>  981                 break;
>>>>>  982         case -3:
>>>>>  983                 tst_brkm(TCONF, cleanup_fn, "requires a NUMA system.");
>>>>>
>>>>> it always failed with "requires a NUMA system.".
>>>>>
>>>>> I think a NUMA system may have only one node, right? If so, could we
>>>>> change the count to 1?
>>>>
>>>> Hmm, are you sure that the testcases are actually applicable to a system
>>>> with only one node?
>>>>
>>>> Because what the testcases do in the setup is to write 1 to
>>>> /sys/kernel/mm/ksm/merge_across_nodes so I would expect that the test
>>>> needs at least two nodes...
>>>>
>>> I think if there is only one NUMA node, then this is a special case for
>>> this testcase, it'll test if set "merge_across_nodes" to 1, any problem
>>> will be triggered.
>>>
>>> I have tried to change the node count to 1 in get_allowed_nodes() and
>>> recompiled ksm02. It works successfully.
>>>
>>
>>
>> Do you think ONLY ONE NODE can be called NUMA?
>>
>> http://en.wikipedia.org/wiki/Non-uniform_memory_access
>>
> 
> I think it's not a problem to run the testcase on a non-NUMA system, if
> we take it as a special case?

It's meaningless, the "special case" you said is already there, it's ksm01.

Thanks,
Wanlong Gao

> 
> Thanks.
> 
> 
> ------------------------------------------------------------------------------
> Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
> from Actuate! Instantly Supercharge Your Business Reports and Dashboards
> with Interactivity, Sharing, Native Excel Exports, App Integration & more
> Get technology previously reserved for billion-dollar corporations, FREE
> http://pubads.g.doubleclick.net/gampad/clk?id=157005751&iu=/4140/ostg.clktrk
> _______________________________________________
> Ltp-list mailing list
> Ltp-list@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/ltp-list
> 


------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=157005751&iu=/4140/ostg.clktrk
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] NUMA system and get_a_numa_node()

[Han Pingtian]

On Thu, Nov 27, 2014 at 04:11:15PM +0800, Wanlong Gao wrote:
> On 11/27/2014 03:45 PM, Han Pingtian wrote:
> >>
> >>
> >> Do you think ONLY ONE NODE can be called NUMA?
> >>
> >> http://en.wikipedia.org/wiki/Non-uniform_memory_access
> >>
> > 
> > I think it's not a problem to run the testcase on a non-NUMA system, if
> > we take it as a special case?
> 
> It's meaningless, the "special case" you said is already there, it's ksm01.
> 

All right. I hope I can figure out what features is ksm02 testing.

But looks like ksm02 tests "set_mempolicy()" and "cpuset", which ksm01
doesn't deal with. Isn't there any meanings to test them on a single
node system?

Thanks.


------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=157005751&iu=/4140/ostg.clktrk
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list