Improved Read Speeds

[zain]

Greetings. I was just wondering if there was a way to improve the read speeds of the array so that they are on par with what you would get with 6 disks in a raidz2/raid6 setup? I am currently testing Unraid on a 30 day trial and this performance, or lack thereof, is really hindering my desire to keep this OS. I do have a couple of SSDs in the cache pool but that isn't going to give me the overall performance boost that I need.

 

This question comes as a last-ditch effort to keep Unraid and not have to migrate everything over to a new OS <again>.

 

Thanks!

[JorgeB]

Unraid doesn't stripe data in the array, so read speed can never be faster than a single disk, pools can be faster.

[zain]

Thanks Jorge, so I gather Unraid isn’t quite the right OS for my use case. Thanks!

[zain]

Is it acceptable to use Pools only (no array devices)?

 

My thinking here is that if I can put all of my drives in a pool and utilize raid 6 through the pool, I would have the read benefits that I'm looking for. So basically I'd have a pool of 6 drives in Raid 6 instead of having an array of 6 drives. But then is there now no cache ability utilizing SSDs?

[JonathanM]

26 minutes ago, zain said:

Is it acceptable to use Pools only (no array devices)?

Currently you must have 1 data disk assigned, but that can be a USB flash drive (not the boot drive).

[zain]

But this would probably also mean the ability to have ssd caching would go away, correct?

Edited March 8, 2021 by zain

[JonathanM]

2 hours ago, zain said:

But this would probably also mean the ability to have ssd caching would go away, correct?

Not sure what you mean by this.

 

Unraid cache pools are used in 2 ways, primarily now days they are used as a higher speed location for docker containers and their data, and VM vdisks. Originally they were used as a temporary write destination for files that would be moved to the parity protected array overnight, as writing directly to the parity array is much slower. Now with modern drives and equipment, you can come reasonably close to saturating a gigabit ethernet connection writing directly to the parity array in write reconstruct (turbo) mode, so using the cache pool as a temporary destination isn't as prevalent as it once was.

 

Writes to user shares either go directly to one of the parity array drives, or one of the cache pools, depending on the setting for the user share. Then when the mover script is scheduled, the files may be moved from a cache pool to the parity array, once again depending on the settings. Reading a file ALWAYS happens directly from the drive where the file is located, and overwriting a file always overwrites to the same drive. Reads happen at the speed of the media, either a single disk in the parity array, or the BTRFS RAID level defined on the pool where the file is located. Writes to new and existing files on the parity array are restricted to roughly 1/3 or 1/4 the raw speed of the disk because of parity calculation and writes, this is improved to roughly 1/2 raw media speed with turbo mode.

[zain]

Thanks Jon. I understand what you have posted above however I’m running a 10gb network here between server and workstation. Read speeds and data protection are important to me. The read speed of a single spinning drive is not enough. That’s why I’m looking to setup an array (or pool?) in raid6 or 10.

[JorgeB]

You can use raid5/6 btrfs pools for better performance, I have several, just be aware that there are still some known issues with btrfs and raid5/6, so it's important to have good backups.

[zain]

Thanks Jorge. I think I’m going to migrate to a different system with RAID support. 

[MDN]

Am I correct in assuming that there is another way to improve reading speed? I'm thinking of putting in 2+ SSDs and then forming a share from just those SSDs. With a share you can specify which hard disks should be included or excluded. Normal SATA SSDs have a read speed of over 500MB/s. That is up to 5 times faster than many HDDs. If you have 2 or more SSDs, the data is distributed over several hard disks and this sub array can be expanded at any time and is protected via the parity hard disk.

 

@zain I would be interested to know which NAS you have chosen now.

[zain]

TrueNAS SCALE. I was on the released CORE version but have recently switched to the Beta release of SCALE. 

 

Unraid has its advantages but if you want the benefits of true raid, look at other NAS operating systems.

 

I have a 10g internal network for data transfer speeds and wanted the read/write speed gains of stripes mirrors (raid 10).

[MDN]

Yes, TrueNAS was also in my selection. It has some advantages for professional use. The big advantage of unRaid for me and my small budget is that you can easily expand the unRaid array with single disks. This is not possible with TrueNAS and ZFS. This way I can buy and install a new hard drive every now and then.

[zain]

Yes. That is one of those things to have to factor in when deciding which OS you want. TrueNAS is free, too.

 

I liked Unraid but it didn’t suit my needs. With that being said, I have a backup server I’ll be moving off site and plan to use Unraid for it. 
 

they all have their quirks perks and drawbacks, you just have to weigh them out when deciding. lol I moved all my data several times while figuring it out!

[gumby327]

I think ... (I am a systems developer)  there is more value in the dockers themselves like Plex Server to build out a file read cache.  On demand when a movie is called for it starts writing it out to a NVMe or even a memory ram disk.  Then when (like last night) your wife is watching Deadpool, and you are watching Real Player One, both are on Disk4 and both are high IO.  One of you will buffer and the buffer is related to the count of arms in your hard drive (one).

 

It is my belief that reading contiguously from disk to any sort of a ram disk or NVMe would be faster than a typical LAN bandwidth.  So, I would like to see the Plex developers create a setting for pre-cache locations.

Edited January 16, 2022 by gumby327