Re: [LTP] [PATCH resend 5/6] mm: use new numa_helper

From: Zhouping Liu <zliu@redhat.com>
To: Jan Stancek <jstancek@redhat.com>
Cc: LTP List <ltp-list@lists.sourceforge.net>
Subject: Re: [LTP] [PATCH resend 5/6] mm: use new numa_helper
Date: Fri, 10 Aug 2012 05:54:38 -0400 (EDT)	[thread overview]
Message-ID: <23193633.36569632.1344592478934.JavaMail.root@redhat.com> (raw)
In-Reply-To: <2014107378.3027931.1344587065940.JavaMail.root@redhat.com>

----- Original Message -----
> From: "Jan Stancek" <jstancek@redhat.com>
> To: "Zhouping Liu" <zliu@redhat.com>
> Cc: "LTP List" <ltp-list@lists.sourceforge.net>, "Caspar Zhang" <caspar@casparzhang.com>
> Sent: Friday, August 10, 2012 4:24:25 PM
> Subject: Re: [LTP] [PATCH resend 5/6] mm: use new numa_helper
> 
> 
> 
> ----- Original Message -----
> > From: "Zhouping Liu" <zliu@redhat.com>
> > To: "Caspar Zhang" <caspar@casparzhang.com>
> > Cc: "LTP List" <ltp-list@lists.sourceforge.net>
> > Sent: Friday, 10 August, 2012 10:02:01 AM
> > Subject: Re: [LTP] [PATCH resend 5/6] mm: use new numa_helper
> > 
> 
> > > -» return·nnodes;
> > > +» ret·=·get_allowed_nodes(NH_MEMS|NH_CPUS,·1,·&nd1);
> > 
> > Caspar, I found get_a_numa_node always return the first node id,
> > which
> > will make all cases called the function
> > only run in the first node, I don't think it's a good way, it
> > decrease
> > the coverage.
> 
> We had a discussion with Caspar on this topic:
> http://article.gmane.org/gmane.linux.ltp/16495
> 
> Can you explain why a specific node matters here?

yeah, I didn't carefully read the discussion before,
and discussed with Caspar, I agree with the current solution.
so Reviewed-by: Zhouping Liu <zliu@redhat.com>

but always I'm worrying it maybe has difference between different nodes,
especially in old version, such as RHEL5, which didn't improve the
performance of scheduling in NUMA system, so that more process will choose
the first node(I guess), and in a large NUMA system, which have 2+ available
NUMA nodes(contain memory and CPUs), I think the second node represent more
universality than first one, that's all my supposing, so I will investigating
the supposing, actually if it exists the difference between different nodes,
I will do a new patch.

Thanks,
Zhouping 

> 
> Regards,
> Jan
> 
> > 
> > we should prefer to run cases on the second node, we can chose the
> > first
> > only if it just only has
> > one node with NH_MEMS|NH_CPUS. do you think so?
> > 
> > and others is good for me.
> > 
> > Thanks,
> > Zhouping
> 
>

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list