Author Topic: Transparent RAID Performance Thread - Part 1/4 (S.W.O)  (Read 16098 times)

Offline Brahim

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 8,402
  • Karma: +201/-15
    • View Profile
Transparent RAID Performance Thread - Part 1/4 (S.W.O)
« on: August 13, 2013, 10:30:55 pm »
Transparent RAID is designed with data integrity and resiliency at the forefront.
Nevertheless, it is also designed to provide performance that is much greater than similar offerings.

As it is typical in RAID, greater performance can be achieved by giving up a bit on resiliency (data integrity should never be compromised upon, but resiliency can be traded off).
In tRAID, however, our primary focus will be on performance tweaks that are fully resilient even in the event of system crash or power loss.
Later on, we will focus on tweaks that are relatively safe for systems that are trusted stable and with proper UPS (some resiliency offloaded).

The first two performance tweaks enabled since RC6 are resiliency safe. That is, they will not compromise the RAID even in the event of system crash or power loss.

  • Direct I/O vs Buffered I/O:
    The default setting is for Direct I/O. Some systems might perform better using Buffered I/O (setting Direct I/O to false).
    Most systems should see no performance difference between the two settings.
     
  • Sequential Write Optimization (S.W.O):
    This setting improves sequential writes, but might degrade random writes.
    You need to evaluate your particular usage to determine whether you care for improved sequential writes or not.
    The default settings have this optimization enabled since most workloads tend to be heavy on sequential write performance than random writes.
    You should play with the optimization level to see what works best for you. The higher the optimization, the greater the memory usage and the greater the impact on random writes. Setting it too high might also have a negative effect sequential writes themselves. So, finding the optimal setting is key.
« Last Edit: March 02, 2014, 08:34:45 am by Brahim »

Offline Brahim

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 8,402
  • Karma: +201/-15
    • View Profile
Re: Transparent RAID Performance Thread - Part 1
« Reply #1 on: August 14, 2013, 08:09:14 pm »
Here are some comparison results for one of my setups for reference.

Test setup as follow:
- 2x Seagate Barracuda 7200.14 ST3000DM001 3TB 7200 (one of which is the PPU)
- 2x Samsung HD204UI 5400 RPM 2TB
- AMD Athlon X2 BE-2350 Brisbane 2.1GHz Socket AM2
- Biostar AM2 mATX mobo with 5 SATA ports
- Windows 8 64 bit
- 4GB RAM

1. Normal speed for the disk tested outside of tRAID and with Windows caching off (this is really important) is 70MB/s writes.

2. I get a steady 40-45MB/s writes with S.W.O set to 8 or ~65% of normal speed

3. I get a steady 30-32MB/s with S.W.O off (set to 0) or ~46% of normal speed

When testing the source disk speed, it is important to test with write caching off to get the true speed and not the speed influenced by Windows' caching.
Alternatively, copy a file large enough to exhaust Windows' cache to get the true speed of the drive.
Essentially, copy a file that is 10GB or larger and note the speed at which the copy settles before it ends.

Moreover, caching will be enabled in the future on the tRAID disks once we are all comfortable with things, which will boost speeds even further.
I will need users to make sure their systems are properly stable and that they have a good UPS before caching is enabled.


Offline Brahim

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 8,402
  • Karma: +201/-15
    • View Profile
Re: Transparent RAID Performance Thread - Part 1
« Reply #2 on: August 14, 2013, 08:20:55 pm »
Also, note that you will get higher transfer rates if you increase the SWO value, but you speed might end up fluctuating greatly, which is a sign of it being set too high.

So, the higher transfer rate is not the goal. The goal is the highest relatively steady transfer rate.
Speed will always fluctuate. You just don't want to much of a fluctuation.

Offline vletroye

  • Hero Member
  • *****
  • Posts: 714
  • Karma: +7/-0
    • View Profile
Re: Transparent RAID Performance Thread - Part 1/3
« Reply #3 on: October 20, 2013, 02:15:59 pm »
I am experiencing quite low performances when writing into the pool (only 30MB/s)


So I will restart measuring the perf of my disks outside tRaid, with Windows Caching disabled like this (right ?) :

1 ) Go to "Control Panel\System and Security\Administrative Tools"
2 ) Open the "Computer Management"
3 ) Click on "Storage" > "Disk Management"
4 ) Right-click the drive on which you want to enable or disable disk write caching, and then click Properties.
6 ) Click the "Policies" tab.
7 ) Click to select or clear "Enable Write Caching on the device"
8 ) Click OK.

Just to be sure there is nothing else that this setting.
« Last Edit: October 20, 2013, 02:36:29 pm by vletroye »

Offline Brahim

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 8,402
  • Karma: +201/-15
    • View Profile
Re: Transparent RAID Performance Thread - Part 1/3
« Reply #4 on: October 20, 2013, 09:47:28 pm »
9) Reboot.

Offline vletroye

  • Hero Member
  • *****
  • Posts: 714
  • Karma: +7/-0
    • View Profile
Re: Transparent RAID Performance Thread - Part 1/3
« Reply #5 on: October 23, 2013, 04:15:02 pm »
This going to possibly impact the speed of the write operations of data on the DRU.
But does it impact also the speed of the "parity creation" process ?

V.

Offline terrastrife

  • Full Member
  • ***
  • Posts: 235
  • Karma: +7/-0
  • 38TB FlexRAID
    • View Profile
Re: Transparent RAID Performance Thread - Part 1/3
« Reply #6 on: October 23, 2013, 05:16:50 pm »
I am experiencing quite low performances when writing into the pool (only 30MB/s)
This is pretty normal if you are at the end of your disk, the end is always slower and thrashing both read and write at the same time means slower rates.
If you have a new 4TB 7200RPM disk as parity I would expect better performance, say a minimum of 50MB/sec but if not, start tweaking your cache settings.
30MB/sec is the minimum I write at with a 5900RPM Parity disk, it's closer to 70MB/sec at the start.

Offline vletroye

  • Hero Member
  • *****
  • Posts: 714
  • Karma: +7/-0
    • View Profile
Re: Transparent RAID Performance Thread - Part 1/3
« Reply #7 on: October 24, 2013, 03:44:45 pm »
This is a bit off topic as IO and SWO are not concerned.

So I have created a new thread to continue further on "how to improve my write speed": http://forum.flexraid.com/index.php/topic,2859.0.html

V.