Jump to content
RESET Forums (homeservershow.com)

Dear Microsoft: Storage Spaces - Get It Right This Time


JackoUK
 Share

Recommended Posts

Looking at the following Arstechnica post again I have certain demands of Microsoft for Storage Spaces.
 
It is NOT ACCEPTABLE for writes to mirrored storage spaces to take twice as long: I'm expecting maybe 105% of a single disk write.
 
And it's NOT ACCEPTABLE for reads from a single column mirrored space to take the same time as a single disk: they should take 55%.
 
And as the column count in a mirrored space goes up, I'm expecting a near linear performance increase in reading.
 
The architecture I want for disks is exactly the same as SMB multichannel in the networking realm:
- the equivalent of a landing zone (or tiering to operate in that fashion)
- disk writes should happen in parallel with multiple columns
- disk reads should happen in parallel with multiple columns
Writing of finished projects from a workstation to a file server is thus:
- SSD to 4NIC's out
- 4NIC's in to a landing zone SSD in the file server
- landing zone to 4-column mirror at near single disk speed
That should provide a throughput of apprximately 400MBs. With consumer grade components.
Reading should have a similar throughput but will not go through the file server SSD since all columns operate simultaneously.
 
And another demand. Can we have some good documentation please? And some sample commandlets. And how about some test scripts to check our installation is working? And how about some guidelines on column sizing; block size and interleave?
 
Microsoft's performance with home storage has been uniformly crap for many years:
- WinFS was abandoned
- WHS had some nice features (and a few bugs) but was abandoned
- Windows Server Essentials 2012 R2 at home? Yeah thrust some more monetising nonsense using your exisitng technology down our throats instead of doing the proper thing.
- Storage Spaces. Crap in rev 1.
- OneDrive: still consolidating and tinkering.
 
Look, Microsoft: this is our lifetime data; SMB multichannel looks good on the networking front; OneDrive looks good on the cloud storage front and Storage Spaces looks good on the home storage front and ReFS is coming soon, yes?
 
BUT YOU HAD BETTER MAKE THEM ALL WORK RELIABLY FOREVER THIS TIME.

 

 
Or I for one will be heaping (more) terabytes of complaints on your sorry solutions.

 

RAID was invented in 1988: time for something better?

Link to comment
Share on other sites

A mirror is the equivalent of a single drive. They need to work on their write speed, but read speed is as fast as a single drive. 

What size files are you copying, that has you looking for 400MB network throughput?

Link to comment
Share on other sites

But, in a mirror, read speed should be faster, due to "multi-channel" reading. The system should start reading from one drive, but also have the other drive prefetching and caching the next block of data. Technically, the drives aren't striped, but reads should take place almost as though they are.

Link to comment
Share on other sites


Your article is from 2012...... Storage Spaces has been improving -- I'm looking forward to experimenting with the newer version once the new beta of Server 2016 is out. Most of my experimenting was with S2012 & S2012E and just a little with S2012R2E Preview.  With the early versions I found transfer rates improved with more physical drives added to the pool -- all other things being equal.

http://homeservershow.com/creating-and-measuring-the-performance-of-a-storage-space-pool-in-windows-server-2012-essentials-r2-preview-part-2.html

 

It holds a lot of promise IMO.

 

If you have some more recent experimental results in Storage Spaces please be sure to post.....

Edited by Joe_Miner
Link to comment
Share on other sites

But, in a mirror, read speed should be faster, due to "multi-channel" reading. The system should start reading from one drive, but also have the other drive prefetching and caching the next block of data. Technically, the drives aren't striped, but reads should take place almost as though they are.

With RAID controllers you will see reads from mirrored drives be a little faster, but not much. Theoretically, if you are able to switch back and forth between the two drives and you are reading from the drives cache it could be faster. The problem is, you read from the cache much faster than the cache refills, so after a couple of drive switches you would be back to the limit of internal transfer rates. Small files may be faster, but large files would be limited by the transfer rate of a single drive. Do you know of anyone doing multi-channel reading? 

Link to comment
Share on other sites

"Do you know of anyone doing multi-channel reading?"

My naive (ignorant) expectation is that a consumer-grade 6Gbs HBA (be it RAID capable or not) can process all 4 SATA disks connected to it simultaneously, if so directed.

Should this not be the case, then I would need an HBA per disk to process data in parallel ... or a fancy HBA of some sort.

 

"If you have some more recent experimental results in Storage Spaces ..."

I don't but this guy does:

http://www.miru.ch/why-column-size-does-matter-with-storage-spaces/

 

 

Link to comment
Share on other sites

Do you know of anyone doing multi-channel reading? 

 

I'm pretty sure my 3ware 9650SE-4LPML does it, and Wikipedia has this:

 

RAID 1 consists of an exact copy (or mirror) of a set of data on two or more disks; a classic RAID 1 mirrored pair contains two disks. This layout is useful when read performance or reliability is more important than the resulting data storage capacity; such an array can only be as big as the smallest member disk.[5][6]

Depending on the nature of I/O load, random read performance of a RAID 1 array may equal up to the sum of each member's performance,[a] while the write performance remains at the level of a single disk.

 

I haven't used a RAID1 array in quite a while, but I do recall that reading from a mirror was faster than from a single. I think this became most obvious when I had to break a mirror one time and I noticed a slowdown after rebooting with only a single disk. Now, I'm not going to claim that the read performance was cut in half with a single disk, but I recall that the difference was noticeable, enough that I wondered what was going on until I did a facepalm and said, "Duh, of course".

Link to comment
Share on other sites

-snip-

 

"If you have some more recent experimental results in Storage Spaces ..."

I don't but this guy does:

http://www.miru.ch/why-column-size-does-matter-with-storage-spaces/

 

 

 

 

Not much opportunity to use a 24 bay in the home but an interesting overview.  If you want some additional reading here's the SS Survival Guide: http://blogs.technet.com/b/josebda/archive/2014/11/19/storage-spaces-survival-guide-links-to-presentations-articles-blogs-tools.aspx

 

In http://homeservershow.com/storage-spaces-performance-in-windows-server-2012-essentials-on-a-hp-n40l-proliant-microserver.htmlI had done some testing with 5 drive SS summarized in Table 1:

 

gallery_1229_27_57873.jpg

 

Using NTFS and Thin Provision I found I could get pretty decent performance with a 2-way Mirror.  Others have found similar performance.

 

You can find a number of blog postings on SS at http://homeservershow.com/tag/storage-spaces

Link to comment
Share on other sites

Bear with me please while I try to catch up ... you guys have 3+ years head start!

 

I've found Joe's test results very useful. I'll post future tests under the storage spaces section but offer the following conjectures in the meantime.

 

1. The ReFS results stand out ... we might call it the polarised file system ... so good are the read figures in comparison to the dismal writes.

 

2. Single disk max read was 342 without storage spaces, Virtual disk H: with ReFS 5-disk 2-way mirror was 1504. This surely answers the question about parallel reads from multiple SATA disks on a single HBA, representing a near linear increase in performance per disk.

 

3. The write performance also appears to be linear ... but inverted ... single disk max of 191, ReFS 5-disk 2-way mirror 50: which is about 191/5 ish.

 

4. It looks like ReFS has to write to disks in a serial manner ... but once in place can read in a parallel manner. We could thus do with a landing zone configuration where data to be written was cached quickly to an SSD and then written (slowly) to ReFS.

 

5. ReFS is not currently available in Windows client (except via a registry hack) I believe ... but soon perhaps. Of all the NTFS configurations only the simple volume 700 v 342 offers compelling performance gains. Perhaps then a configuration in two parts consisting of a write to a simple volume, followed by a plain synchronisation to a copy. (Note to self: must check how the write back cache in the tiering feature works.)

 

6. It would be useful to see the effect of scaling for 3,4,5,6,7,8 disks. Lots of work ... guess I need to start doing some of it! I think 8 disks is well within an enthusiast PC build.

 

7. I’ve picked up another beneficial design feature of Storage Spaces with regard to recovering from disk failures. When a disk on one side of a mirror fails then:

- rebuilding begins immediately to the other disks in the mirror (providing they have capacity)

- to ALL the other disks simultaneously, much better than a hot-spare which would be limited to one disk and otherwise idle and not participating in general mirror performance

I’m beginning to think Microsoft is going the right way here.

 

So my wishlist for Storage Spaces v 2.0 on Windows 10:

a. Disk usage rebalancing.

b. Option to use ReFS.

c. Landing zone feature (if that cannot be handled by tiering)

Then I think Microsoft might really have something.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share

×
×
  • Create New...