You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@nifi.apache.org by "Takashi Menjo (Jira)" <ji...@apache.org> on 2021/02/12 02:57:00 UTC

[jira] [Updated] (NIFI-8219) PMEM-backed Repositories

     [ https://issues.apache.org/jira/browse/NIFI-8219?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Takashi Menjo updated NIFI-8219:
--------------------------------
    Description: 
*Please note that this patchset works only on x64 Linux (4.15 or later) for now.*

Persistent memory (PMEM) is non-volatile and byte-addressable memory installed into DIMM slots. With Filesystem DAX (Direct Access) and PMDK (Persistent Memory Development Kit), a program can map on-PMEM files to userspace then read and write its data, bypassing page caches. These technologies could bring you better I/O performance than traditional disks.

I would propose a patchset that lets FlowFile, Content, and Provenance Repositories use PMDK (via JNI) to write their data to their files shown as follows:
 * FlowFile Repository: Journals (.journal)
 * Content Repository: Content/Resource Claims
 * Provenance Repository: Provenance logs (.prov) and ToCs (.toc)

Prerequisites on compile: gcc, libpmem-devel (1.5 or later; 1.9 or later recommended)

Prerequisites on run: libpmem (1.5 or later; 1.9 or later recommended)

  was:
Persistent memory (PMEM) is non-volatile and byte-addressable memory installed into DIMM slots. With Filesystem DAX (Direct Access) and PMDK (Persistent Memory Development Kit), a program can map on-PMEM files to userspace then read and write its data, bypassing page caches. These technologies could bring you better I/O performance than traditional disks.

I would propose a patchset that lets FlowFile, Content, and Provenance Repositories use PMDK (via JNI) to write their data to their files shown as follows:

 * FlowFile Repository: Journals (.journal)
 * Content Repository: Content/Resource Claims
 * Provenance Repository: Provenance logs (.prov) and ToCs (.toc)

*Please note that this patchset works only on x64 Linux (4.15 or later) for now.*


> PMEM-backed Repositories
> ------------------------
>
>                 Key: NIFI-8219
>                 URL: https://issues.apache.org/jira/browse/NIFI-8219
>             Project: Apache NiFi
>          Issue Type: New Feature
>          Components: Core Framework
>         Environment: PMEM, x64, Linux, DAX, PMDK
>            Reporter: Takashi Menjo
>            Priority: Minor
>
> *Please note that this patchset works only on x64 Linux (4.15 or later) for now.*
> Persistent memory (PMEM) is non-volatile and byte-addressable memory installed into DIMM slots. With Filesystem DAX (Direct Access) and PMDK (Persistent Memory Development Kit), a program can map on-PMEM files to userspace then read and write its data, bypassing page caches. These technologies could bring you better I/O performance than traditional disks.
> I would propose a patchset that lets FlowFile, Content, and Provenance Repositories use PMDK (via JNI) to write their data to their files shown as follows:
>  * FlowFile Repository: Journals (.journal)
>  * Content Repository: Content/Resource Claims
>  * Provenance Repository: Provenance logs (.prov) and ToCs (.toc)
> Prerequisites on compile: gcc, libpmem-devel (1.5 or later; 1.9 or later recommended)
> Prerequisites on run: libpmem (1.5 or later; 1.9 or later recommended)



--
This message was sent by Atlassian Jira
(v8.3.4#803005)