Review request (S): 7179517 Enable NUMA by default on NUMA hardware

[Jesper Wilhelmsson]

I have updated the webrev with a Solaris patch as well. It's the same URL, 
just do a reload in your browser.

I'm not a Solaris native either, so regarding testing I have asked the 
performance team to look extra close on Solaris to make sure it behaves as 
expected.
/Jesper


On 2012-07-02 19:26, Vladimir Kozlov wrote:
> That question was not for you but for Jesper.
>
> Vladimir
>
> Eric Caspole wrote:
>> I am hoping one of you guys can do it - I have never used Solaris and I
>> don't know if there is any special thing to take into account there.
>>
>>
>> On Jul 2, 2012, at 12:07 PM, Vladimir Kozlov wrote:
>>
>>> Why there is no changes for Solaris?
>>>
>>> Vladimir
>>>
>>> On 7/2/12 3:57 AM, Jesper Wilhelmsson wrote:
>>>> Hi,
>>>>
>>>> This change is to turn on UseNUMA per default on NUMA hardware. It has
>>>> been contributed by Eric Caspole. I'm fine with
>>>> the change but we need at least one official reviewer.
>>>>
>>>> Webrev:
>>>> http://cr.openjdk.java.net/~jwilhelm/7179517/webrev/
>>>>
>>>> Testing:
>>>> I ran it through JPRT and did some local sanity testing.
>>>> Eric has tested this on SPECjbb2005 and SPECjvm2008 with a variety of heap
>>>> sizes and it is never worse than the current
>>>> default. In some cases, especially with Windows, running a test with the
>>>> current default and getting bad NUMA placement
>>>> will be about 3x slower than the +UseNUMA score.
>>>>
>>>> The performance team are involved to perform some larger scale testing.
>>>>
>>>
>>
>>