Shulou(Shulou.com)06/01 Report--

Apply PMDK to modify WAL operation to fit persistent memory

These patches can read and write WAL logs stored in persistent memory PMEM by using PMDK. PMEM is the next generation storage medium, which has a series of characteristics: fast, byte addressing and non-volatile.

Pgbench is the general benchmark of PG, tested with benchmark, and the PG modified by these patches improves the performance of the native PG by 5%. With our insert benchmark, it can be 90% faster than the native PG. The following is described in detail.

This e-mail includes the following parts:

A) PMDK

B) Patch

C) Test methods and results

PMDK

PMDK provides functions that allow applications to access PMEM directly without having to use the kernel as memory. API includes:

1) API of open PMEM file, API of create PMEM file, API of map PMEM file to virtual address

PMDK takes advantage of the DAX file system feature to provide these API functions. The DAX file system is sensitive to PMEM and allows direct access to PMEM without using the kernel's page cache. You can use the standard mmap function to map files in the DAX file system to memory. Further, by mapping files in PMEM to virtual addresses, applications can use CPU's load/store instruction instead of read/write to access PMEM.

2) API for reading and writing PMEM files

PMDK provides API: similar to the memcpy () function, which copies data to PMEM through single instruction, multiple data instruction, and NT storage instruction. These instructions can improve copy performance. So these API are faster than read/write. API reference:

[1] http://pmem.io/pmdk/

[2] https://www.usenix.org/system/files/login/articles/login_summer17_07_rudoff.pdf

[3] SIMD: a single instruction that operates on loading data. If the SIMD system loads 8 bytes of data into registers at a time, then the storage operation to PMEM will be performed on all 8-byte values at the same time.

[4] NT store instructions: this instruction skips CPU cache, so flush is not required to use this instruction.

Patch

Patch modification:

0001-Add-configure-option-for-PMDK.patch: add-- with-libpmem configuration, execute IO through the pmdk library

0002-Read-write-WAL-files-using-PMDK.patch:

IO WAL using the PMDK function

The parameter wal_sync_method is added with pmem-drain to indicate the wal sync mode on the PMEM.

0003-Walreceiver-WAL-IO-using-PMDK.patch:

For the walreciver process of the slave, use PMDK to log.

Implementation mode and result

Environment:

Server: HP ProLiant DL360 Gen9

CPU: Xeon E5-2667 v4 (3.20GHz); 2 processors (without HT)

DRAM: DDR4-2400; 32 GiB/processor

(8GiB/socket x 4 sockets/processor) x 2 processors

NVDIMM: DDR4-2133; 32 GiB/processor

(8GiB/socket x 4 sockets/processor) x 2 processors

HDD: Seagate Constellation2 2.5inch SATA 3.0. 6Gb/s 1TB 7200rpm x 1

OS: Ubuntu 16.04, linux-4.12

DAX FS: ext4

NVML: master (at) Aug 30, 2017

PostgreSQL: master

Note: I bound the postgres processes to one NUMA node, and the benchmarks to other NUMA node.

1) configure pmem as a block device

# ndctl list

# ndctl create-namespace-f-e namespace0.0-- mode=memory-M dev

2) create a file system on pmem and mount it in DAX mode

# mkfs.ext4 / dev/pmem0

# mount-t ext4-o dax / dev/pmem0 / mnt/pmem0

3) set PMEM_IS_PMEM_FORCE, indicating that the WAL file is stored on the PMEM

Note that if this environment variable is not set, the PG process cannot be started

# export PMEM_IS_PMEM_FORCE=1

4) install PG

There are three important considerations when installing Pg:

A. add-- with-libpmem: ". / configure-- with-libpmem" when Configure

b. Store the WAL directory on PMEM

c. Change the wal_sync_method parameter from fdatasync to pmem_drain

Specific operations:

# cd / path/to/ [PG _ source dir]

#. / configure-with-libpmem

# make & & make install

# initdb / path/to/PG_DATA-X / mnt/pmem0/path/to/ [PG _ WAL dir]

# cat / path/to/PG_DATA/postgresql.conf | sed-es / # wal_sync_method\ =\

Fsync/wal_sync_method\ =\ pmem_drain/ > / path/to/PG_DATA/postgresql.conf.

Tmp

# mv / path/to/PG_DATA/postgresql.conf.tmp / path/to/PG_DATA/postgresql.conf

# pg_ctl start-D / path/to/PG_DATA

# created [DB_NAME]

5) execute 2 benchmark, one is pgbench and the other is my insert benchmark

Pgbench:

# numactl-N1 pgbech-c 32-j 8-T 120-M prepared [DB_NAME]

The average of three times performing the pgbench:

Wal_sync_method=fdatasync: tps = 43179

Wal_sync_method=pmem_drain: tps = 45254

Pclinet_thread:my insert benchmark

Prepare:

CREATE TABLE [TABLE_NAME] (id int8, value text)

ALTER TABLE [TABLE_NAME] ALTER value SET STORAGE external

PREPARE insert_sql (int8) AS INSERT INTO% s (id, value) values ($1,'

[1K_data]')

Execute:

BEGIN; EXECUTE insert_sql (% lld); COMMIT

Note: I ran this quer 5M times with 32 threads.

#. / pclient_thread

Invalid Arguments:

Usage:. / pclient_thread [The number of threads] [The number to insert

Tuples] [data size (KB)]

# numactl-N 1. / pclient_thread 32 5242880 1

Average of three tests:

Wal_sync_method=fdatasync: tps = 67780

Wal_sync_method=pmem_drain: tps = 131962

Attachment

Content-Type

Size

0001-Add-configure-option-for-PMDK.patch

Application/octet-stream

5.1 KB

0002-Read-write-WAL-files-using-PMDK.patch

Application/octet-stream

46.9 KB

0003-Walreceiver-WAL-IO-using-PMDK.patch

Application/octet-stream

4.8 KB

Original text

Https://www.postgresql.org/message-id/C20D38E97BCB33DAD59E3A1%40lab.ntt.co.jp

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.