![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Windows 8 Preview / I am playing with Storage Spaces
- Thread starter Silenus
- Start date
Trepidati0n
[H]F Junkie
- Joined
- Oct 26, 2004
- Messages
- 9,270
As of now, Flexraid is not real time. However, rumors are someday it might.
I think 2.0 Beta does have a "real time" configuration in it. I only need snapshot anyway as I'll only be changing my data once a week. I'll just have the parity snapshot run nightly.
Trepidati0n
[H]F Junkie
- Joined
- Oct 26, 2004
- Messages
- 9,270
My issue with that method is I don't think FlexRAID works with ReFS. Secondarily, I only have had dealings with FlexRAID on a failed drive.
From what I've read, FlexRaid seems like a vastly superior solution to the problem I think 99% of people looking for an advanced storage solution are trying to solve, which is - "I want a easy, space effective, safe and efficient solution for my multi-TB data, spread across multiple disks of varying sizes".
By definition, most of this data is write once. The main requirements boil down to -
- easy to add disks, easy to maintain
- performance
- efficiency. only spin up the disk needed
- reliable
- safe. data should be in native format on the fs
- cheap. in product cost (no hw raid) and disk cost (must be parity based)
At present, this is fulfilled only by unRaid and FlexRaid on Windows. Copy your data, generate a check sum at regular intervals, and FlexRaid works great. The developer has decided to make it commerical and is also working on a very ambitious ZFS like product.
SS/ReFS offer nothing to me. None of the advanced options are exposed in the UI, the product itself is designed to meet very different needs, and I don't think this is going to change. MS took > 6 months to fix a data corruption bug in WHS v1, they abandoned WinFS after years of development, they removed DE from WHS v2 due to technical difficulties, I don't have much confidence in them solving this problem. MS do have excellent dedup which they use in their server SIS (and also in WHS backups) but its a shame this is not part of SS or ReFS.
By definition, most of this data is write once. The main requirements boil down to -
- easy to add disks, easy to maintain
- performance
- efficiency. only spin up the disk needed
- reliable
- safe. data should be in native format on the fs
- cheap. in product cost (no hw raid) and disk cost (must be parity based)
At present, this is fulfilled only by unRaid and FlexRaid on Windows. Copy your data, generate a check sum at regular intervals, and FlexRaid works great. The developer has decided to make it commerical and is also working on a very ambitious ZFS like product.
SS/ReFS offer nothing to me. None of the advanced options are exposed in the UI, the product itself is designed to meet very different needs, and I don't think this is going to change. MS took > 6 months to fix a data corruption bug in WHS v1, they abandoned WinFS after years of development, they removed DE from WHS v2 due to technical difficulties, I don't have much confidence in them solving this problem. MS do have excellent dedup which they use in their server SIS (and also in WHS backups) but its a shame this is not part of SS or ReFS.
On-disk post-processing dedupe is baked into the core OS now, it functions on ReFS & NTFS. Of course the question then becomes at what edition point does it become enabled.
I dream of the day we'll get a FreeNAS that will have a 1 disk Parity function in ZFS (for obvious data integrity).
Unraid/Flexraid are by far the best options out there for "Multimedia Data Storage".
I understand why MS did this and its awesome.. but... STRIPED DATA ? seriously....
All they had to do was N+1 Drives with the +1 as Parity.
Take 2% of every drive's capacity and store the entire array's file metadata there.
D1 => F1 F2 F3 F4 + List of Fs
D2 => F5 F6 F7 F8 + List of Fs
D3 => F9 F10 F11 + List of Fs
Parity => PARITY + List of Fs
Using the "List of Fs Metadata" it shouldn't be too hard to make a virtual pool (hell flexraid did it).
While providing protection against a "single drive failure" and also "no array loss if more than 1 failure" since no striping.
Having a distributed List of Fs also gives you a 100% guaranteed visibility of "what is stored where" and "what files were lost if 2+ failures".
I'm SHOCKED they actually used striping on a "home" product.
They using stripping as thier mirrored set is actually a RAID10 setup which is smart enough to work on two drives. As you add more drives, Windows 8 is apparently smart enough to only use the least used drive which has a copy of the data you need.
There are good solid reasons todo that, and ReFS (which in a decade or so) should eventually replace NTFS has very good integration with Storage Spaces.
There are good solid reasons todo that, and ReFS (which in a decade or so) should eventually replace NTFS has very good integration with Storage Spaces.
Major bug/limitation with SS - http://arstechnica.com/civis/viewtopic.php?f=15&t=1168727
ZFS is really easy and intuitive. No need for 20 commands to create a raid, and format and setup partitions etc etc.
Create a zfs raid:
# zpool create myRaidNow raidz2 disk0 disk1 disk2 disk3 disk4
this creates a raid6 (raidz2 is raid6). No need to format, just start to copy data immediately. To create raid-5 (raidz1)
# zpool create myRaidNow raidz1 disk0 disk1 disk2 disk3 disk4
To create a single disk:
# zpool create mySingleDisk disk0
Create a new partition (similar to create D: or E
# zfs create myRaidNow/HereBeMyMovies
this creates a new partition, no need to format. Just start to move data. The zfs partition grows and shrinks as needed. (In normal Linux/Unix filesystems, the partition is fixed in size and does not grow/shrink as needed).
On top of that, you can add an gui. For instance, napp-it. No need to learn Unix or Linux. I never use any gui for zfs, it is easier to type commands as above, than learn a web gui such as napp-it:
http://www.eviltyrant.com/blog/?p=988
There is a big difference in manageability. MSFT has worked with enclosure manufacturers to enable blinking of drive LED's from a GUI console, and a lot of other things that make it easier to manage that ZFS on Solaris. Can you say drive labels that don't change when you move a disk to a different slot?
Also, the biggest problem ZFS has is that I can't incrementally add a disk to an existing raidz array - I have to move the data to other storage, tear down the array, and then rebuild it and move the data back. Almost all hardware controllers support this kind of expansion, and so does md on linux, and Windows 8 software raid does as well, with thin provisioning to boot.
Given that Oracle is not adding features like this to ZFS, this is a big opportunity for MSFT.
I am a unix hack and run Linux software raid and windows hardware raid at home. I was thinking about converting to ZFS, but if MSFT can fix performance on writes, that's where I am going to move to. It's easier to manage and easy to expand storage, which are my two primary drivers in selection.
This is fundementally impossible with sata over sas. A sata drive simply doesn't have the unique ID which is identifiable over the disk subsystem that a sas drive has, and no amount of smarts with how ZFS/Solaris deals with this can solve it. The only solution is to write that unique ID onto the disk into the filesystem headers/descriptors, which is how hardware controllers which do array roaming do it.
I had some updates for Win8 customer preview this morning, installed them. I have 5 X 2TB Samsung disks in parity. Sending the Win8 ISO will go at over 100MB/sec for about half the transfer, then it throttles down to between 25-50MB/sec for the rest.
I have an Intel Pro1000 PT NIC installed with Intel drivers.
Sorry for the crappy upload pic.
http://i41.tinypic.com/5wcjk9.jpg
Jim
I have an Intel Pro1000 PT NIC installed with Intel drivers.
Sorry for the crappy upload pic.
http://i41.tinypic.com/5wcjk9.jpg
Jim
brutalizer
[H]ard|Gawd
- Joined
- Oct 23, 2010
- Messages
- 1,600
You can blink a drive lamp by writing a Unix command. You ask zfs to read from the disk you specify, to /dev/null. Then only that single disk will be active, and the led will be active.
You can not grow a zfs raid by adding a single disk, instead you add several disks as a group. A zfs raid consists of several groups of disks (I think that a large hardware raid also consists of several groups of disks?).
The main reason many use zfs, is because of data corruption protection. That is not easy to do well. As CERN said "adding checksums will not necessarily provide data protection". I dont use zfs to get high performance. What is the best performance in the world worth, if your data might be corrupt?
drescherjm
[H]F Junkie
- Joined
- Nov 19, 2008
- Messages
- 14,937
In linux I find drives in hot swap raid cages that have leds using dd.
Something like the following will show me where /dev/sdm is:
Something like the following will show me where /dev/sdm is:
Code:
dd if=/dev/sdm of=/dev/null bs=8M count=100I see a lot of "... but Unix/Linux/Solaris can do this as well, just do..." followed by a series of commands that I am not interested in running, and seemingly don't have to run if I choose a Windows solution. I've seen this before, commonly in non-zfs threads on here.
I like Microsoft's offering. It essentially gives me more flexibility than I had previously with WHS, with more data reliability, and I am still in a familiar Windows environment. I'm still sad at the abysmal parity write speeds, but I can find ways around this.
najames, your speeds are interesting. I wonder what's causing that weird speed drop-off.
I like Microsoft's offering. It essentially gives me more flexibility than I had previously with WHS, with more data reliability, and I am still in a familiar Windows environment. I'm still sad at the abysmal parity write speeds, but I can find ways around this.
najames, your speeds are interesting. I wonder what's causing that weird speed drop-off.
I don't know for sure, but I suspect it is Windows reducing speed. This pattern and speed seems very consistent, I've done it several times. If I use the same Win7 "sending PC" to a different Win7 "receiving PC" (instead of the Win8 PC), the transfer is over 100MB/sec the entire transfer. In WIn8 I tried changing the NetworkThrottlingIndex in the registry, but it makes no difference if set to 0 or 200, default is 10.
You are familiar with Windows, but Win8 is a different critter. I'd like it a lot better if they still had the option to hack the registry and return to traditional Windows operation, not a fan of Metro, deleted almost all the crap they had on Metro. The saving feature is being able to pin things on the bottom bar on the desktop, just missing a easy power button as it is currently.
If I wasn't using this box for a SageTV server (already have Windows licenses), and just wanted a file server, I'd probably use Linux.
You're probably still better off installing FlexRaid.
MS solution would be perfect IF (and its a big IF) they did not strip data.
Raid5/6 in for business makes sense (since there will be a 24/7 monitoring, hot spares, etc etc).
But for a Home usage ? lose 1 drive, be careless= lose everything.
Last edited:
Multiple parity drives. With Raid-6 or Raidz2/z3 and reasonably sized pools (8-10) drives your chance of actually loosing data are pretty miniscule (you are probably more likely to be robbed).
As for monitoring - an email letting you know there is a problem should be supported by just about everything as is pretty straightforward.
I'd note that there are other alternatives to your rather scant list, for example snapraid.
http://snapraid.sourceforge.net/
In my opinion, the developer of FlexRAID is a genius but he doesn't seem to be able to finish one version before starting another, or nowadays starting a new product. This makes me wary of buying into a commercial FlexRAID - at least until there is a 2.x version.
I have done that too. It works when the filesystem is not under heavy load, but it's really dumb to have to push data to the drive just to light an LED. And command line typing sometimes gets fat fingered and causes problems. As I said, I am a unix hack, and am very comfortable with the command line. That said, one day I fat fingered a command and wiped out a volume accidentally because I was in a hurry and screwed up of and if on a dd.
We should expect better tools than this. It is 2012 already.
But in this case, for home use it would be better to have duplicate copies of data to protect against drive failure, which just turns things into a RAID 1 anyway, which Storage Spaces can also handle.
If you're just talking about write-once, read-many data made up of movies that is not mission-critical (at least not compared to non-replaceable business information) then a non-striped, parity redundancy scheme provides the best balance of risk vs. usable space for the home user. If you mirror you lose half your usable space. I'd use parity for most things at home. My documents are the only thing that warrant an extra safeguard against loss because they are irreplaceable--but I use dropbox for that so I don't need to bother with mirroring my docs.
As it stands now, Storage Spaces is a big leap in the right direction but the striping of data is going to be a dealbreaker for some. You lose 2 drives (assuming 1 parity drive) and you lose everything. Unnecessary risk/hassle for media storage.
As it stands now, Storage Spaces is a big leap in the right direction but the striping of data is going to be a dealbreaker for some. You lose 2 drives (assuming 1 parity drive) and you lose everything. Unnecessary risk/hassle for media storage.
Isn't this the risk for any RAID 5-like solution?
I understand snapshot RAID (like FlexRAID) gets around this by the nature of how it works, but you don't get the read performance boost of a live RAID system. But I can see now more why it's preferred; 2 or more drives can fail without loosing 100% data, and you don't have the write performance loss as with Storage Spaces.
Yeah, raid 5 stripes the data for increased performance but also means your individual files are spread across multiple hard drives and require the raid functioning to see them. Flexraid and Unraid basically remove the striping element of the striping+parity that makes up Raid 5.
The other benefit of removing striping is, with Flexraid and Unraid, you can remove an HDD at any time and pop it onto a Windows machine and see your files. With Flexraid, I believe you can also set it up on drives already containing data. Write performance is still bad, compared to a real raid 5 (I believe unraid gets around 30-40 MB/sec without a cache drive), but the point with these systems is that they are perfect for home media storage; you don't need the write speeds of raid 5 and the attendant extra risks that come with it.
The other benefit of removing striping is, with Flexraid and Unraid, you can remove an HDD at any time and pop it onto a Windows machine and see your files. With Flexraid, I believe you can also set it up on drives already containing data. Write performance is still bad, compared to a real raid 5 (I believe unraid gets around 30-40 MB/sec without a cache drive), but the point with these systems is that they are perfect for home media storage; you don't need the write speeds of raid 5 and the attendant extra risks that come with it.
Raid-5 is a red herring. You should be running Raid-6/Raid-Z2 with large 2TB + disks anyway.
If we are only talking 2-3TB of data then sure, just mirror it with two drives - but assuming we are talking multi disk arrays just run 2 parity disks and you will be fine.
If we are only talking 2-3TB of data then sure, just mirror it with two drives - but assuming we are talking multi disk arrays just run 2 parity disks and you will be fine.
drescherjm
[H]F Junkie
- Joined
- Nov 19, 2008
- Messages
- 14,937
It would be much better to make backup than RAID1 (or any other raid level) and store the backup off site or at minimum not powered on all the time.
This doesn't protect against a single drive failure, as this is not live. Storage Spaces shouldn't replace a sound backup strategy anymore than a real RAID solution.
drescherjm
[H]F Junkie
- Joined
- Nov 19, 2008
- Messages
- 14,937
delete
I have written up many of the points discussed in this thread in the following blog post:
Windows 8 Storage Spaces: Bugs and Design Flaws
Windows 8 Storage Spaces: Bugs and Design Flaws
While I like what you have posted in your blog, and I commend how clean and concise it makes the Storage Spaces discussion, I have to point out that you're essentially saying Storage Spaces main problem is that it isn't Windows Home Server v1 Drive Extender. All of the architecture faults along with the issues of disk space utilization and balancing seem already "fixed" in the first version of DE, and I remember the same points came up when discussing Microsoft DEv2 for WHS 2011 (which was ultimately removed and placed here).
But most of those problems I still see with any regular RAID system as well; I have to use same-size drives, I can't yank a drive out and read it directly, adding a drive kills data on said drive, 2+ drive failures kill the array, etc. These seem less like faults and more like concessions for using anything that stripes data, and none of your solutions have the improved read speed that parity RAID systems give.
At this point, it looks like Storage Spaces only adds slightly more functionality to regular old Windows Software RAID.
But most of those problems I still see with any regular RAID system as well; I have to use same-size drives, I can't yank a drive out and read it directly, adding a drive kills data on said drive, 2+ drive failures kill the array, etc. These seem less like faults and more like concessions for using anything that stripes data, and none of your solutions have the improved read speed that parity RAID systems give.
At this point, it looks like Storage Spaces only adds slightly more functionality to regular old Windows Software RAID.
brutalizer
[H]ard|Gawd
- Joined
- Oct 23, 2010
- Messages
- 1,600
I think you are bit too harsh. Many of the shortcomings also applies to ZFS, but still people considers ZFS as good. So many of the shortcomings you point out, are not really catastrophic, you can work your way around that. For instance, when you add new disk, the data is not rebalanced. Well, just create a new folder, and move your data to the new folder which means the data will be rebalanced.
Some storage spaces vs. RAID testing:
http://blog.insanegenius.com/2012/09/22/storage-spaces-leaves-me-empty/
http://blog.insanegenius.com/2012/09/22/storage-spaces-leaves-me-empty/
http://thedocsworld.net/server-2012-storage-spaces-testing-the-rtm-version/
Another review.
It just doesn't look good for SS.
Another review.
It just doesn't look good for SS.