Very poor write performance in RAID 5 configuration using Intel Rapid Storage Technology (IRST)

Aug 5, 2018
4
0
10
I configured 5 2TB Seagate SATA disks using the Intel Rapid Storage Technology in a RAID 5 configuration.

I am pleased with the Read performance but the write performance is very poor. Seagate Specs. show 210 MB/s max sustained transfer rate on a single disk. I realize there is some overhead with the RAID 5 configuration but 31.85 MB/s sequential write is too low.

Does anyone know how to improve the write performance of my RAID 5 array.

Thank you for your assistance.

Benchmark results:

-----------------------------------------------------------------------
CrystalDiskMark 6.0.1 x64 (UWP) (C) 2007-2018 hiyohiyo
Crystal Dew World : https://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes

Sequential Read (Q= 32,T= 1) : 751.966 MB/s
Sequential Write (Q= 32,T= 1) : 31.850 MB/s
Random Read 4KiB (Q= 8,T= 8) : 8.338 MB/s [ 2035.6 IOPS]
Random Write 4KiB (Q= 8,T= 8) : 2.071 MB/s [ 505.6 IOPS]
Random Read 4KiB (Q= 32,T= 1) : 7.498 MB/s [ 1830.6 IOPS]
Random Write 4KiB (Q= 32,T= 1) : 1.466 MB/s [ 357.9 IOPS]
Random Read 4KiB (Q= 1,T= 1) : 1.383 MB/s [ 337.6 IOPS]
Random Write 4KiB (Q= 1,T= 1) : 1.321 MB/s [ 322.5 IOPS]

Test : 1024 MiB [D: 0.0% (0.7/7452.0 GiB)] (x5) [Interval=5 sec]
Date : 2018/08/26 17:35:29
OS : Windows 10 Professional [10.0 Build 17134] (x64)

System Information:

OS name: Microsoft Windows 10 Pro
OS version: 10.0.17134
Processor: GenuineIntel Intel64 Family 6 Model 158 Stepping 10 3.696 GHz
BIOS: American Megatrends Inc., 1602
PCH: 0xA2C9

Intel® Rapid Storage Technology enterprise Information
Kit installed: 16.0.2.1086
User interface version: 16.0.2.1086
Language: English (United States)
RAID option ROM version: 16.0.2.3402
Driver version: 16.0.2.1086
ISDI version: 16.0.2.1086

Storage System Information

RAID Configuration:

Array Name: SATA_Array_0000
Size: 9,538,644 MB
Available space: 9 MB
Number of volumes: 1
Volume member: RAID5
Number of array disks: 5
Array disk: Z4ZBPG4Q
Array disk: Z4ZBKW1K
Array disk: Z4ZBQ351
Array disk: Z4ZBWYCL
Array disk: Z4ZBPF6K
Disk data cache: Enabled

Volume name: RAID5
Status: Normal
Type: RAID 5
Size: 7,630,907 MB
System volume: No
Data stripe size: 128 KB
Write-back cache: Write back
Initialized: Yes
Parity errors: 0
Blocks with media errors: 0
Physical sector size: 4096 Bytes
Logical sector size: 512 Bytes

Hardware Information:

Controller name: Intel(R) Chipset SATA/PCIe RST Premium Controller \\Scsi1
Type: SATA
Mode: RAID
Number of SATA ports: 6
Number of volumes: 1
Volume: RAID5
Number of spares: 0
Number of available disks: 1
Rebuild on Hot Plug: Disabled
Manufacturer: Intel Corporation
Model number: 0x2822
Product revision: 0
Direct attached disk: Z4ZBPG4Q
Direct attached disk: Z4ZBKW1K
Direct attached disk: Z4ZBQ351
Direct attached disk: S3Z8NB0K535996Y
Direct attached disk: Z4ZBWYCL
Direct attached disk: Z4ZBPF6K


Disk on Controller 0, (Used 5 identical 2TB disks in the RAID 5 configuration)
Status: Normal
Type: SATA disk
Location type: Internal
Usage: Array disk
Size: 1,863 GB
System disk: No
Disk data cache: Enabled
Command queuing: NCQ
Transfer rate: 6 Gb/s
Model: ST2000DM006-2DM164
SCSI device ID: 0
Firmware: CC26
Physical sector size: 4096 Bytes
Logical sector size: 512 Bytes

Disk on Controller 0, Port 3 (Samsung SSD disk for OS)
Status: Normal
Type: SATA SSD
Location type: Internal
Usage: Available
Size: 932 GB
System disk: Yes
Disk data cache: Enabled
Command queuing: NCQ
Transfer rate: 6 Gb/s
Model: Samsung SSD 860 EVO 1TB
Serial number: S3Z8NB0K535996Y
SCSI device ID: 0
Firmware: RVT01B6Q
Physical sector size: 512 Bytes
Logical sector size: 512 Bytes

The rest of the hardware configuration:

CPU: Intel - Core i7-8700K 3.7GHz 6-Core Processor
CPU Cooler: Cooler Master - Hyper 212 LED Turbo (Red) 66.3 CFM CPU Cooler
Motherboard: Asus - ROG MAXIMUS X HERO (WI-FI AC) ATX LGA1151 Motherboard
Memory Corsair - Vengeance RGB 32GB (2 x 16GB) DDR4-3000 Memory
Storage SSD Samsung - 860 Evo 1TB 2.5" Solid State Drive
Storage SATA (RAID 5) 5 Seagate - Barracuda 2TB 3.5" 7200RPM Internal Hard Drive
Video Card Asus - GeForce GTX 1070 8GB Video Card
Case Cooler Master - MasterCase Pro 5 ATX Mid Tower Case
Power Supply Corsair - 850W 80+ Platinum Certified Fully-Modular ATX Power Supply
Optical Drive Asus - DRW-24B1ST/BLK/B/AS DVD/CD Writer
SATA PCIe Card Vantec 2 Channel 4-Port SATA 6Gbps PCIe Host Card (UGT-ST622)
 
Solution
sharam, after some research (sorry for the delay in responding) there are a few things to consider when measuring RAID 5 performance. At least as you enabled write back cache and NCQ so the write performance at the drive level shouldn't be throttled.

When you mentioned 210MB/Sec sequential data rate, that is the throughput measured at the outer diameter of the disk. Towards the inner diameter of the disk, the number will drop down to about 1/2 of that. Depending on which part of the disk the performance benchmark runs, the numbers can vary from 105MB/Sec to 210MB/Sec on a single drive level.

As for the RAID 5 write performance goes, the maximum expected performance should be 262.5MB/Sec (number of HDDs*max throughput/4). It...
sharam, after some research (sorry for the delay in responding) there are a few things to consider when measuring RAID 5 performance. At least as you enabled write back cache and NCQ so the write performance at the drive level shouldn't be throttled.

When you mentioned 210MB/Sec sequential data rate, that is the throughput measured at the outer diameter of the disk. Towards the inner diameter of the disk, the number will drop down to about 1/2 of that. Depending on which part of the disk the performance benchmark runs, the numbers can vary from 105MB/Sec to 210MB/Sec on a single drive level.

As for the RAID 5 write performance goes, the maximum expected performance should be 262.5MB/Sec (number of HDDs*max throughput/4). It could be as low as 131.25MB/Sec as the inner diameter of the disk. However, this performance figure assumes that there is no system overhead. IRST is one of those software RAID that adds a lot of overhead as the the system has to read data from multiple drives to calculate parity data before writing. I don't have actual figures from this particular set up but all the blogs I have read indicate some horrific write performance numbers.

Some suggestions I would make are:
1. Adjust RAID stripe size. Generally larger stripes give you better numbers.
2. Use hardware RAID controller. LSI and Marvell make RAID controllers which will give you the performance figures mentioned above.

Hope this helps.
 
Solution