[ping] Re: [11] RFR(M): 8189922: UseNUMA memory interleaving vs membind

David Holmes david.holmes at oracle.com
Tue Jul 10 21:39:04 UTC 2018


Hi Gustavo,

On 11/07/2018 6:14 AM, Gustavo Romero wrote:
> Hi Swati,
> 
> As David pointed out, it's necessary to determine if that bug qualifies 
> as P3 in order to get it into JDK 11 RDP1.
> 
> AFAICS, that bug was never triaged explicitly and got its current 
> priority (P4) from the default.

Actually no, the P4 was from the (Oracle internal) ILW prioritization 
scheme.

For this to be a P3 it needs to be shown either that the impact is quite 
significant (IIUC it's only a mild performance issue based on the bug 
report); or that the likelihood of this being encountered is very high 
(again it seems not that likely based on the info in the bug report).

HTH.

David
-----


> 
> Once it's defined the correct integration version, I can sponsor that 
> change
> for you. I think there won't be any updates for JDK 11 (contrary to what
> happened for JDK 10), but I think we can understand how distros are 
> handling
> it and so find out if there is a possibility to get the change into the
> distros once it's pushed to JDK 12.
> 
> 
> David, Alan,
> 
> I could not find a documentation on how to formally triage a bug. For 
> instance,
> on [1] I see Alan used some markers as "ILW =" and "MLH = " but I don't 
> know if
> these markers are only for Oracle internal control. Do you know how could I
> triage that bug? I understand its risk of integration is small but even 
> tho I
> think it's necessary to bring up additional information on that to 
> combine in a
> final bug priority.
> 
> Thanks.
> 
> 
> Best regards,
> Gustavo
> 
> [1] https://bugs.openjdk.java.net/browse/JDK-8206953
> 
> On 07/03/2018 03:06 AM, David Holmes wrote:
>> Looks fine.
>>
>> Thanks,
>> David
>>
>> On 3/07/2018 3:08 PM, Swati Sharma wrote:
>>> Hi David,
>>>
>>> I have added NULL check for _numa_bitmask_isbitset in 
>>> isbound_to_single_node() method.
>>>
>>> Hosted:http://cr.openjdk.java.net/~gromero/8189922/v2/ 
>>> <http://cr.openjdk.java.net/~gromero/8189922/v2/>
>>>
>>> Swati
>>>
>>> On Mon, Jul 2, 2018 at 5:54 AM, David Holmes <david.holmes at oracle.com 
>>> <mailto:david.holmes at oracle.com>> wrote:
>>>
>>>     Hi Swati,
>>>
>>>     I took a look at this though I'm not familiar with the functional
>>>     operation of the NUMA API's - I'm relying on Gustavo and Derek to
>>>     spot any actual usage errors there.
>>>
>>>     In isbound_to_single_node() there is no NULL check for
>>>     _numa_bitmask_isbitset (which seems to be the normal pattern for
>>>     using all of these function pointers).
>>>
>>>     Otherwise this seems fine.
>>>
>>>     Thanks,
>>>     David
>>>
>>>
>>>     On 30/06/2018 2:46 AM, Swati Sharma wrote:
>>>
>>>            Hi,
>>>
>>>         Could I get a review for this change that affects the JVM when
>>>         there are
>>>         pinned memory nodes please?
>>>
>>>         It's already reviewed and tested on PPC64 and on AARCH64 by
>>>         Gustavo and
>>>         Derek, however both are not Reviewers so I need additional
>>>         reviews for that
>>>         change.
>>>
>>>
>>>         Thanks in advance.
>>>
>>>         Swati
>>>
>>>         On Tue, Jun 19, 2018 at 5:58 PM, Swati Sharma
>>>         <swatibits14 at gmail.com <mailto:swatibits14 at gmail.com>> wrote:
>>>
>>>             Hi All,
>>>
>>>             Here is the numa information of the system :
>>>             swati at java-diesel1:~$ numactl -H
>>>             available: 8 nodes (0-7)
>>>             node 0 cpus: 0 1 2 3 4 5 6 7 64 65 66 67 68 69 70 71
>>>             node 0 size: 64386 MB
>>>             node 0 free: 64134 MB
>>>             node 1 cpus: 8 9 10 11 12 13 14 15 72 73 74 75 76 77 78 79
>>>             node 1 size: 64509 MB
>>>             node 1 free: 64232 MB
>>>             node 2 cpus: 16 17 18 19 20 21 22 23 80 81 82 83 84 85 86 87
>>>             node 2 size: 64509 MB
>>>             node 2 free: 64215 MB
>>>             node 3 cpus: 24 25 26 27 28 29 30 31 88 89 90 91 92 93 94 95
>>>             node 3 size: 64509 MB
>>>             node 3 free: 64157 MB
>>>             node 4 cpus: 32 33 34 35 36 37 38 39 96 97 98 99 100 101 
>>> 102 103
>>>             node 4 size: 64509 MB
>>>             node 4 free: 64336 MB
>>>             node 5 cpus: 40 41 42 43 44 45 46 47 104 105 106 107 108 109
>>>             110 111
>>>             node 5 size: 64509 MB
>>>             node 5 free: 64352 MB
>>>             node 6 cpus: 48 49 50 51 52 53 54 55 112 113 114 115 116 117
>>>             118 119
>>>             node 6 size: 64509 MB
>>>             node 6 free: 64359 MB
>>>             node 7 cpus: 56 57 58 59 60 61 62 63 120 121 122 123 124 125
>>>             126 127
>>>             node 7 size: 64508 MB
>>>             node 7 free: 64350 MB
>>>             node distances:
>>>             node   0   1   2   3   4   5   6   7
>>>                 0:  10  16  16  16  32  32  32  32
>>>                 1:  16  10  16  16  32  32  32  32
>>>                 2:  16  16  10  16  32  32  32  32
>>>                 3:  16  16  16  10  32  32  32  32
>>>                 4:  32  32  32  32  10  16  16  16
>>>                 5:  32  32  32  32  16  10  16  16
>>>                 6:  32  32  32  32  16  16  10  16
>>>                 7:  32  32  32  32  16  16  16  10
>>>
>>>             Thanks,
>>>             Swati
>>>
>>>             On Tue, Jun 19, 2018 at 12:00 AM, Gustavo Romero <
>>>             gromero at linux.vnet.ibm.com
>>>             <mailto:gromero at linux.vnet.ibm.com>> wrote:
>>>
>>>                 Hi Swati,
>>>
>>>                 On 06/16/2018 02:52 PM, Swati Sharma wrote:
>>>
>>>                     Hi All,
>>>
>>>                     This is my first patch,I would appreciate if anyone
>>>                     can review the fix:
>>>
>>>                     Bug :
>>>                     https://bugs.openjdk.java.net/browse/JDK-8189922
>>>                     <https://bugs.openjdk.java.net/browse/JDK-8189922> <
>>>                     https://bugs.openjdk.java.net/browse/JDK-8189922
>>>                     <https://bugs.openjdk.java.net/browse/JDK-8189922>>
>>>                     Webrev
>>>                     :http://cr.openjdk.java.net/~gromero/8189922/v1
>>>                     <http://cr.openjdk.java.net/~gromero/8189922/v1>
>>>
>>>                     The bug is about JVM flag UseNUMA which bypasses the
>>>                     user specified
>>>                     numactl --membind option and divides the whole heap
>>>                     in lgrps according to
>>>                     available numa nodes.
>>>
>>>                     The proposed solution is to disable UseNUMA if bound
>>>                     to single numa
>>>                     node. In case more than one numa node binding,
>>>                     create the lgrps according
>>>                     to bound nodes.If there is no binding, then JVM will
>>>                     divide the whole heap
>>>                     based on the number of NUMA nodes available on the
>>>                     system.
>>>
>>>                     I appreciate Gustavo's help for fixing the thread
>>>                     allocation based on
>>>                     numa distance for membind which was a dangling issue
>>>                     associated with main
>>>                     patch.
>>>
>>>
>>>                 Thanks. I have no further comments on it. LGTM.
>>>
>>>
>>>                 Best regards,
>>>                 Gustavo
>>>
>>>                 PS: Please, provide numactl -H information when
>>>                 possible. It helps to
>>>                 grasp
>>>                 promptly the actual NUMA topology in question :)
>>>
>>>                 Tested the fix by running specjbb2015 composite workload
>>>                 on 8 NUMA node
>>>
>>>                     system.
>>>                     Case 1 : Single NUMA node bind
>>>                     numactl --cpunodebind=0 --membind=0 java -Xmx24g
>>>                     -Xms24g -Xmn22g
>>>                     -XX:+UseNUMA
>>>                     -Xlog:gc*=debug:file=gc.log:time,uptimemillis
>>>                     <composite_application>
>>>                     Before Patch: gc.log
>>>                     eden space 22511616K(22GB), 12% used
>>>                            lgrp 0 space 2813952K, 100% used
>>>                            lgrp 1 space 2813952K, 0% used
>>>                            lgrp 2 space 2813952K, 0% used
>>>                            lgrp 3 space 2813952K, 0% used
>>>                            lgrp 4 space 2813952K, 0% used
>>>                            lgrp 5 space 2813952K, 0% used
>>>                            lgrp 6 space 2813952K, 0% used
>>>                            lgrp 7 space 2813952K, 0% used
>>>                     After Patch : gc.log
>>>                     eden space 46718976K(45GB), 99% used(NUMA disabled)
>>>
>>>                     Case 2 : Multiple NUMA node bind
>>>                     numactl --cpunodebind=0,7 –membind=0,7 java -Xms50g
>>>                     -Xmx50g -Xmn45g
>>>                     -XX:+UseNUMA
>>>                     -Xlog:gc*=debug:file=gc.log:time,uptimemillis
>>>                     <composite_application>
>>>                     Before Patch :gc.log
>>>                     eden space 46718976K, 6% used
>>>                            lgrp 0 space 5838848K, 14% used
>>>                            lgrp 1 space 5838848K, 0% used
>>>                            lgrp 2 space 5838848K, 0% used
>>>                            lgrp 3 space 5838848K, 0% used
>>>                            lgrp 4 space 5838848K, 0% used
>>>                            lgrp 5 space 5838848K, 0% used
>>>                            lgrp 6 space 5838848K, 0% used
>>>                            lgrp 7 space 5847040K, 35% used
>>>                     After Patch : gc.log
>>>                     eden space 46718976K(45GB), 99% used
>>>                             lgrp 0 space 23359488K(23.5GB), 100% used
>>>                             lgrp 7 space 23359488K(23.5GB), 99% used
>>>
>>>
>>>                     Note: The proposed solution is only for numactl
>>>                     membind option.The fix
>>>                     is not for --cpunodebind and localalloc which is a
>>>                     separate bug bug
>>>                     https://bugs.openjdk.java.net/browse/JDK-8205051
>>>                     <https://bugs.openjdk.java.net/browse/JDK-8205051>
>>>                     and fix is in progress
>>>                     on this.
>>>
>>>                     Thanks,
>>>                     Swati Sharma
>>>                     Software Engineer -2 at AMD
>>>
>>>
>>>
>>>
>>>
>>
> 


More information about the hotspot-dev mailing list