jdk and NUMA

charlie hunt charlie.hunt at oracle.com
Thu Sep 9 19:05:15 PDT 2010


David Dabbs wrote:
>> Charlie Hunt
>> In general, best practice 
>> use +UseNUMA when the jvm you are launching will span more than one memory node.
>>     
>
> Hi Charlie.
>
> Apologies for my ignorance, but how would I know that this is the case? 
> We're running a servlet container (Tomcat) with no special CPU affinities or numactl.
> So I'm guessing that this qualifies as spanning more than one memory node
>   
Assuming you have a NUMA system, and you are not using any Linux 
commands to control CPU affinity on the JVM running Tomcat, then yes, 
this would qualify for spanning more than one memory node.

hths,

charlie ...

> Thanks!
>
> David
>
>
>
> -----Original Message-----
> From: [mailto:charlie.hunt at sun.com] 
> Sent: Thursday, September 09, 2010 6:34 PM
> To: David Dabbs; hotspot-dev at openjdk.java.net
> Subject: RE: jdk and NUMA
>
> In general, best practice .... use +UseNUMA when the jvm you are launching will span more than one memory node.
>
> Charlie 
>
> David Dabbs <dmdabbs at gmail.com> wrote:
>
>   
>> Thank you ramki, for the reply.
>>
>> What "issues" should might one expect after adding +UseNUMA?
>>
>>
>>
>>
>> -----Original Message-----
>> From: Y. S. Ramakrishna [mailto: @oracle.com] 
>> Sent: Thursday, September 09, 2010 5:31 PM
>> To: David Dabbs
>> Subject: Re: jdk and NUMA
>>
>> Yes, just use +UseNUMA to begin with, and let us know if there are
>> any issues.
>>
>> -- ramki
>>
>> On 09/09/10 12:04, David Dabbs wrote:
>>     
>>> My apologies if this is not the correct place for this question.
>>>
>>> I have a shiny new NUMA-capable Xeon server and want to ensure that my
>>>       
>> Java
>>     
>>> app(s) make full use of the hardware.
>>> So, I ran java (jdk 6u21) -XX:+PrintFlagsFinal and (naively) expected to
>>>       
>> see
>>     
>>> UseNUMA := true but did not.
>>>
>>> I found this post
>>>
>>> http://blogs.sun.com/jonthecollector/entry/help_for_the_numa_weary
>>>
>>> If its simply a matter of turning it on that's what I'll do. Any
>>>       
>> guidelines
>>     
>>> about use of mumactl or the jdks NUMA "knobs"?
>>>
>>>
>>> uname -a
>>> Linux 2.6.18-128.7.1.el5 #1 SMP Mon Aug 24 08:21:56 EDT 2009 x86_64 x86_64
>>> x86_64 GNU/Linux
>>>
>>> Linux, CentOS 5
>>>
>>>
>>> cat /proc/cpuinfo
>>>
>>> processor       : 0
>>> vendor_id       : GenuineIntel
>>> cpu family      : 6
>>> model           : 26
>>> model name      : Intel(R) Xeon(R) CPU           E5530  @ 2.40GHz
>>> stepping        : 5
>>> cpu MHz         : 2394.112
>>> cache size      : 8192 KB
>>> physical id     : 1
>>> siblings        : 8
>>> core id         : 0
>>> cpu cores       : 4
>>> apicid          : 16
>>> fpu             : yes
>>> fpu_exception   : yes
>>> cpuid level     : 11
>>> wp              : yes
>>> flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca
>>> cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx
>>>       
>> rdtscp
>>     
>>> lm constant_tsc ida nonstop_tsc pni monitor ds_cpl vmx est tm2 cx16 xtpr
>>> popcnt lahf_lm
>>> bogomips        : 4791.64
>>> clflush size    : 64
>>> cache_alignment : 64
>>> address sizes   : 40 bits physical, 48 bits virtual
>>> power management: [8]        
>>>
>>> This repeats for processors 1-15.
>>>
>>> Two physical ids, 0,1.
>>>
>>>
>>>
>>>       
>> No virus found in this incoming message.
>> Checked by AVG - www.avg.com 
>> Version: 9.0.851 / Virus Database: 271.1.1/3124 - Release Date: 09/09/10
>> 01:34:00
>>
>>     
>
> No virus found in this incoming message.
> Checked by AVG - www.avg.com 
> Version: 9.0.851 / Virus Database: 271.1.1/3124 - Release Date: 09/09/10 13:34:00
>
>   



More information about the hotspot-dev mailing list