AVS Forum banner

261 - 280 of 412 Posts

·
Registered
Joined
·
360 Posts
One thing lacking in DrivePool + Snapraid is the nice, proprietary "recycle bin" feature in Flexraid that gives you extra protection against invalid parity should you delete a file until you run the "update".
 

·
Registered
Joined
·
1,237 Posts
Sorry for being obtuse, I just want to make sure I have my head wrapped around using DrivePool + SnapRaid before I switch over from fRaid.

I would set up DrivePool first which will obviously show the pooled drives content as being on a single giant drive. When I then setup SnapRaid, I will not be pointing to the pooled drive but I take it rather to the individual drives mount points or each drives assigned drive letter, correct? All of this was transparent as fRaid handled everything in one package......
You got it!
 

·
Registered
Joined
·
1,237 Posts
Drivepool finally completed balancing my files. What it did was to make sure that no disk was at more than 90% capacity and also that no drives were less than 10% full. That took about 24 hours to complete. Here's a screenshot from just before it completed:



With that out of the way, I disabled all future balancing. Going forward, when I copy new files directly to the pool drive, they will automatically land on the disk with the most free space. So over time, they should all fill up evenly.



And disabled all plugins:



I also made sure duplication was not enabled since snapraid is going to take care of those duties for me:



So now it was time to sync the array for the first time. There were a few system files that showed up from doing the drivepool balancing, but those were ignored since I told snapraid to do so in the config file.



I have 16GB of RAM, but that wasn't nearly enough to prevent swapping to the C: drive during the period right after scanning the 44 data disks. Server was very sluggish during this period, but finally, after about 20 minutes, memory usage started coming back down:



After a few minutes, memory usage settled in at around just under 90% and CPU usage was very reasonable:



So it looks like it will take about 140 hours for the parity to be calculated. Hopefully those drives with S.M.A.R.T. warnings will make it!



My servers sit in the basement, so I hooked up a Y-cam down there so that I can watch the blinking LEDs from my office upstairs. :)



EDIT: snapraid now shows about 120 remaining, so it might only be 5 more days. :D
 

·
Registered
Joined
·
1,237 Posts
I might. I want to see what the write performance is going to be first just copying to the pool. It should be significantly better than tRAID since that solution does the parity calculations real-time.

One concern with a landing disk is how it will deal with torrents in general and especially with files that exceed the size of the SSD. I had to give up on allowing torrents to save directly to my fRAID arrays, which did utilize a landing disk, and instead resorted to using the C: drive, which did not have that much space.

If I get get close to native write performance directly to the disks, and I don't see why I wouldn't, I will probably just do without and never have to worry about the size of files I copy to the pool or any torrent or other related issues.

Btw, ETA to complete the parity is now 77 hours, so it seems to have picked up speed once it settled in.
 

·
Registered
Joined
·
5,045 Posts
...
Btw, ETA to complete the parity is now 77 hours, so it seems to have picked up speed once it settled in.

I had also noticed that the ETA in the beginning of the initial parity calc. process was rather conservative and continued to improve significantly over time.

_____
Axel
 

·
Registered
Joined
·
241 Posts
Got my IBM M1015 flashed/installed and system back up and running. Added a new DRU and transferred about 1.3GB of movies and TV shows onto it that I have been storing on a system temp drive. Update took about 12 hours but went fine. Tried to run a Validation and same exact problem. I guess I'll be migrating to DrivePool + SnapRaid when I have time.....

pclausen, thanks for the info, updates and the pics of the process in action. Will be good to reference back to for when I get started.

habe
 

·
Registered
Joined
·
241 Posts
One more question about SnapRaid.....

On FlexRaid, I disabled recycle bin for each drive before adding to the pool as fRaid has its own built-in recycle bin feature. With SnapRaid, I leave recycle bin for each drive enabled, correct??
 

·
Registered
Joined
·
1,682 Posts
Why SnapRAID and not disParity?
Disparity doesn't allow as many parity drives, which is the reason one would move to software over hardware raid IMO

One more question about SnapRaid.....

On FlexRaid, I disabled recycle bin for each drive before adding to the pool as fRaid has its own built-in recycle bin feature. With SnapRaid, I leave recycle bin for each drive enabled, correct??
Correct

Are you planning on using the Drivepool SSD Optimizer plugin for a landing disk?

I myself disabled ssd optimizer cause I couldn't get it to work like I wanted.

Quick question, why disable drive balancing? Both strategies have their merits. I prefer for a drive to be filled before moving onto the next drive instead of all drives filling equally. My reasoning for this is that I watch a lot of TV series and this would more likely clump the seasons together which means only one disk needs to be spun up.
 

·
Registered
Joined
·
1,237 Posts
I figured it would be safer/better for the drives to not be filled up all the way. When I first switched from hardware raid to software raid, I started out filling my drives one by one, first with my Blu-Ray rips, then the HD-DVDs, the DVD, and finally the TV Shows. So disk 1-10 was all Blu-Ray, disk 11-14 was all HD-DVD, and so on. But then with the pooling feature of tRAID, anything new I added would not always land on the same disk necessarily. Like the .mkv file itself might be on disk 25 and the metadata/subs on disk 36.

And now, after drivepool has moved files from my 99% full disks to others with less space, I'm sure the files for an individual movie or TV show are scattered all over the place.

As far as drives spinning up, unless you happen to watch a movie with subs, and the subs live on another drive, I would think that only one drive would be spinning at a time. For me this is a moot point since my IBM 1015 HBA controllers keep all my 48 drives spinning at all times. Back when I was on hardware raid, my Areca controller would spin them all down after a configurable amount of time. Of course the issue here was that when I wanted to watch something, ALL the drives had to spin back up. I had them set to come in in a staggered fashion with like a 1 second delay between each drive, so it took about 30 seconds before the first movie/show would start, which was annoying.

Some folks are of the opinion that it is better to leave drives on all the time vs. having them spin up and down several times a day. In my case, between the various maintenance tasks (drivepool, snapraid, emby), sickbeard activity, and with 5 emby clients and music streaming by the 4 female members of the family, my drives would be spinning up and down all day long, so I figured I might as well leave everything on all the time.

Btw, I'm now 21% done on the initial snapraid sync. I was going to wait until it was done before starting emby, torrents, sickbeard etc back on, but I caved last night. :D I also noticed that even though I have duplication disabled in drivepool, it still runs a"Duplication" task at regular intervals. Not sure what it's doing since the amount of "Duplicated" data has remained a constant 8.85 MB since the get go.

So overall stats for drivepool are as follows:



The "Duplication" tasks takes several hours to complete. Any ideas why it would keep doing this? And why does it take several hours? I mean, 8.85 MB out of 47TB is what, like 0.00000001% or something, right?
 

·
Registered
Joined
·
3 Posts
Has anyone considered just installing hard drives a without assigning them a drive letter and mount the drive in an empty folder (not sure if this is viable for Linux)? Seems that I have now been using some variant of this for the better part of a two decades, I first ran across the feature in windows98 and have been very happy with it (the failure of a controller card led me an all software solution). The introduction of USB, esata drives etc. have allowed me to move beyond the physical limitations of my cases and motherboards. Add to that port multiplier cards and i have a system with over 20 drives and 32+TB of space.
And let me clear up one thing right away "No I do not have a bunch of external drives laying around." I have 8 drives in a cheap server case from Newegg and have another 13 drives in a 15 bay server case, this case has hard drives only no other components in the case other than controller cards and power supply.

I installed two of Rosewill's esata controller cards in the server giving me 4 esata ports capable of handling port multipliers. I then installed for port multiplier controllers from Addonics in the second server case. I then conncted the two cases togeather via esata cables. Volia cheap meadia sever with 23 bays and another 6 drives of the motherboard of the primary case. And using the afore mentioned hard drive managing technique along with another feature I believe was introduced in windows Vista (hardlinks and junctions) the data from the twenty plus drives (36+ TB) appear to reside in 10mb virtual drive. I am looking at adding some data protection to the mix which led me to this thread was looking at flexraid but now thinking i will keep doing what i'm doing and add snap raid along with stablebit scanner.
 

·
Registered
Joined
·
1,682 Posts
@pclausen, Ok so, there are two ways to do a simple balance. One is to fill a drive till it's full and the other would be to fill all drives equally. If I were to dump 24 episodes onto my array I would like for all of them to hopefully reside on the same disk instead of spread out like balancing option 2 would do.

I spin my drives down just to be a little more green since I normally eat up gobs of power elsewhere (every little bit counts right?). My problem with files being all over the place means wait time as I have to wait for a new drive to spin up when I go to watch the next episode.
@dragonkeeper, that's exactly how I have mine setup. Just make empty folders on my root drive then map the freshly partitioned drives to said folders. Though, having done this also prompted me to put my root drive in a raid 1 config.
 

·
Registered
Joined
·
3,714 Posts
Has anyone considered just installing hard drives a without assigning them a drive letter and mount the drive in an empty folder (not sure if this is viable for Linux)?
It used to be an advanced recommendation for Flexraid -F http://wiki.flexraid.com/2011/09/29/tips-mounting-your-drives-as-folders-on-windows/

I believe @jim2100 junctioned his own read-only storage pool with this method and used snapraid for parity

To clear everything I said above, I just can't see myself being happy with snapshot anymore after using tRAID. Considering I write to the pool at around 90+MB/s without a landing disk and around 300 with a landing disk, I'd classify myself as "very happy" with the performance. There's also the near-0 downtime for anything ever. I registered all the disks and started the pool without calculating anything, then I ran the verify-sync operation with my array online/in-use to create the parity. When a disk fails it just degrades your UoR by 1 and keeps going. Then I can copy everything from the "failed-disk" (aka the virtual disk which has all of it's data available from the parity fairly quickly), shut down, pull out the actual failed disk, and create a new tRAID really only losing the time it takes to restart and setup the new config. One last thing that I like is that I can test any verify mismatch and simply overwrite the parity info if need be. That seems arbitrary, but I eventually had something slip up in my RAID-F parity and it stopped validating/verifying (I turned off every app on the server just to try for a validate). The only solution to that was to keep everything turned off and rebuild the whole parity (which was another good 15hrs of downtime), so I really like that tRAID lets me do everything live without downtime.

Does Snapraid allow accessing the drives while it calculates parity?
 

·
Registered
Joined
·
125 Posts
..fRaid...fRaid...

It's RAID-F and tRAID. RAID-F. ;)

RAID over File System
Transparent RAID.

One thing lacking in DrivePool + Snapraid is the nice, proprietary "recycle bin" feature in Flexraid that gives you extra protection against invalid parity should you delete a file until you run the "update".
This was discussed in another thread but I'd like to point out here also that you can still make use of the FlexRAID Recycle Bin in SnapRAID, if you use FlexRAID pooling instead of DrivePool. So the loss of the virtual recycle bin is not strictly a con for SnapRAID but actually for DrivePool.

I have 16GB of RAM, but that wasn't nearly enough to prevent swapping to the C: drive during the period right after scanning the 44 data disks. Server was very sluggish during this period, but finally, after about 20 minutes, memory usage started coming back down:
It's probably too late now, but a larger block size decreases RAM requirements. It does increase parity overhead but not by a whole lot unless you have a lot of files smaller than the block size. I think parity needs to be re-calculated when changing blocksize so you may just want to buy more RAM!

On FlexRaid, I disabled recycle bin for each drive before adding to the pool as fRaid has its own built-in recycle bin feature. With SnapRaid, I leave recycle bin for each drive enabled, correct??
Yes and No. The Windows Recycle Bin does not operate when files are deleted from over a network. So if that is how you access the files, then it makes no difference. If you are using DrivePool, again the Windows Recycle Bin will not work over the network. However if you still use FlexRAID for pooling, you will still for sure disable the Windows Recycle Bin for all source drives and the pool, and the FlexRAID virtual Recycle Bin *does* work over the network.

The "Duplication" tasks takes several hours to complete. Any ideas why it would keep doing this? And why does it take several hours? I mean, 8.85 MB out of 47TB is what, like 0.00000001% or something, right?
It might be checking for existing duplicates rather than actually creating them? It's probably taking a long time because SnapRAID is hammering your disks.

By any chance... Could you take a screenshot of Resmon or StableBit Scanner with the Performance column enabled, while SnapRAID is syncing? I'm still waiting to see some performance results from other people using HBAs and Expanders...

Has anyone considered just installing hard drives a without assigning them a drive letter and mount the drive in an empty folder (not sure if this is viable for Linux)?
Pretty sure this is what *most* people already do. New users may not yet be familiar with it so you'll probably see a lot of people asking questions where they are still using drive letters.

Does Snapraid allow accessing the drives while it calculates parity?
Yes, absolutely.

To clear everything I said above, I just can't see myself being happy with snapshot anymore after using tRAID.
Does the complete lack of any bitrot protection in tRAID not concern you? You may be able to get the same write performance with DrivePool, if that's a main concern for you. I will agree though that being able to still access the files that existed on a failed drive is an incredible feature... Personally not important enough for me to overlook tRAID downsides.
 

·
Registered
Joined
·
3,714 Posts
Does the complete lack of any bitrot protection in tRAID not concern you? You may be able to get the same write performance with DrivePool, if that's a main concern for you. I will agree though that being able to still access the files that existed on a failed drive is an incredible feature... Personally not important enough for me to overlook tRAID downsides.
It's not like RAID-F gives you any protection anyway. Validate can only tell you that you "might" of had bitrot. Then you have to test and determine for yourself if you had it. Then you can't really fix it without rewriting the data yourself unless it fell into the tiny chance that you noticed it before updating your parity data.

I don't think snapraid or other snapshot raids do a much better job than that anyway, so if I really want protection I'd have to run storage spaces with ReFS (sucks and inflexible) or ZFS/Btrfs (too many other lost server features for me, or I'd have to use a separate box which I don't want to do)
 

·
Registered
Joined
·
125 Posts
It's not like RAID-F gives you any protection anyway. Validate can only tell you that you "might" of had bitrot. Then you have to test and determine for yourself if you had it. Then you can't really fix it without rewriting the data yourself unless it fell into the tiny chance that you noticed it before updating your parity data.

I don't think snapraid or other snapshot raids do a much better job than that anyway, so if I really want protection I'd have to run storage spaces with ReFS (sucks and inflexible) or ZFS/Btrfs (too many other lost server features for me, or I'd have to use a separate box which I don't want to do)
That's definitely true about RAID-F, but SnapRAID does verify everything against checksums before any other processing. It will detect and be able to repair bitrot without any silent propagation into parity. It won't do it "live" like zfs or btrfs, but it's just as thorough.
 

·
Registered
Joined
·
3,714 Posts
That's definitely true about RAID-F, but SnapRAID does verify everything against checksums before any other processing. It will detect and be able to repair bitrot without any silent propagation into parity. It won't do it "live" like zfs or btrfs, but it's just as thorough.
Interesting. What level checksum is used? Have you had it tell you one was off and verified the error yourself?
 

·
Registered
Joined
·
1,237 Posts
It's probably too late now, but a larger block size decreases RAM requirements. It does increase parity overhead but not by a whole lot unless you have a lot of files smaller than the block size. I think parity needs to be re-calculated when changing blocksize so you may just want to buy more RAM!
I still have 12 empty DIMM slots, so yeah, I'll probably go ahead and pick up some additional 4GB modules that match my existing ones. They are Micron Technology FB-DDR2 PC2-6400. My Supermicro X7DWN+ motherboard is a little picky when it comes to RAM, so I want to make sure I get the exact same stuff and them run memtest86 on it for a while.

It might be checking for existing duplicates rather than actually creating them? It's probably taking a long time because SnapRAID is hammering your disks.
I asked Christopher over on the DrivePool support forum, and his response was:

The duplication pass does more than "duplicate" files. It checks the status of the pool and ensure it's set to the proper status. This means that in some cases, that it will delete extra copies, actually.
Additionally, yes, we store some metadata on the pool (about reparse points and other stuff). This is x3 duplication, so that is most likely what you're seeing.
By any chance... Could you take a screenshot of Resmon or StableBit Scanner with the Performance column enabled, while SnapRAID is syncing? I'm still waiting to see some performance results from other people using HBAs and Expanders...
Sure, here you go:



I'm running 2 IBM 1015s (LSI SAS9220-8i) controllers, each with all 8 channels (2 SFF-8087 cables) connected to SAS2 backplanes, so I would have expected it to perform better than it currently is. I know when I ran tRAID, the performance increased significantly when adding the 2nd 4 lane cable to each backplane.



Yes, absolutely.
Very true, but if you modify/add/delete any files, you will of course have to do another sync to pick up those changes. I caught snapraid complain about one such change this morning.



Btw, I halted the sync for a reboot for system updates, and I told me I had 75 file errors.



I plan to run chkdsk on each of my 44 data volumes once the initial sync has completed. StableBit Scanner told me I had 3 drives with damaged file systems. It was able to fix 2 of them running the built in chkdsk. The 3rd one I had to run from an elevated command line, and it did tell me it fixed it, but Scanner still complains about it. I'll have to fiddle with that some more once the sync is done. I also have 3 volumes with corrupt recycle bins that I haven't been able to resolve yet. They date back to when I was running an early version of tRAID.

Here's the current state of my array that I ran just before starting the sync back up after this mornings reboot:



It is picking up on the issue with the damaged file system on data disk 23 (disk 22 physical) I mentioned above. These are all issues that tRAID never told me about and it did sync successfully many times despite this, so I'll definitely fell better once I get all this completed!

Not sure if I want to replace all the drives with S.M.A.R.T. warnings, that's a lot of disks... Maybe I'll at least do the ones with a high count of sector reallocations. I don't think I should worry about the ones that only have spin retry counts, or should I?
 
261 - 280 of 412 Posts
Top