Am I expecting too much or is there a problem - RAID 10

nutcase

nutcase

nutcase

Soldato
Joined
18 Oct 2002
Posts
7,622
Location
SX, unfortunately
Was trying to watch a film earlier and it was stuttering like mad. When I looked, it would seem my wife was going a bit mad downloading and that would explain why it was stuttering. But even at an absolute max of 75mbit download that's only 9.3 MB/sec for the discs to write. Array is 8x 1.2tb 10k hanging off a P822 with 2GB FBWC with the disk being a static VHDX being the only file on it.

Disk.jpg


Look at that delay - all around 3000ms :( If it should be behaving better, where the heck do I start!?

Thanks :)
 
Embarrassingly I don't know - it seems when I increased the disks from 6 to 8 it broke my PRTG monitoring. Got it all set up again, but will have to ask the Missis to repeat what she was doing...
 
You don't have the full screenshot from above? The disk queue length is at the bottom of that screenshot (although it isn't expanded by default).

You might want to try a quick atto or other benchmark while watching the queue length. To me it sounds like you have caching disabled.
 
That looks like it is because there is multiple writes going on at the same time. Is she downloading 8 episodes at once? maybe that is through itunes or something. Some applications will just hog all the disk like that. I bet it is not the throughput but the amount of individual different writes going on. ie its not a throughput issue its a simultaneous write issue. If it is not that it is will be network bandwidth if you are on a 100mbit lan then that download will take up the lan card bandwidth. I see there is says 100mbit on the right.
 
Show you how much I use performance monitor - never even noticed it has the queue! No I don't - it was on the summary tab anyway which doesn't have the queue :(

I've now restored the proper monitoring to the server and ran ATTO - first thing is the shocking performance with ATTO which tells me it's the VHDX is the problem as running ATTO on the host on the same disk the VHDX is stored on gives MUCH better results (sorry no screenshots at the mo).

Queue length when running ATTO never reaches 5 so I don't think that's an issue?

Network wise the wires is 1Gbit for that portion and then there's a hop from my wife's laptop to an Asus Black Knight router - usual connection speed is well over 100mbit so I don't think it's that.
 
It shows on the right in the screenshot disk activity maxed out at 10mbyte a second and the network speed at 100mbit. Even though the network is not maxed out it does indicated the bottleneck is the network speed at 100mbit. Are you sure the nic is not 100mbit?

Do a copy over the network and see if you can get more than 10mbyte a second.
 
anything I don't mind said:
It shows on the right in the screenshot disk activity maxed out at 10mbyte a second and the network speed at 100mbit. Even though the network is not maxed out it does indicated the bottleneck is the network speed at 100mbit. Are you sure the nic is not 100mbit?

Do a copy over the network and see if you can get more than 10mbyte a second.
Click to expand...
The number at the top-right of each little graph adjusts dynamically depending on what the graph is showing, it doesn't represent the absolute maximum.
 
nutcase_1uk said:
Show you how much I use performance monitor - never even noticed it has the queue! No I don't - it was on the summary tab anyway which doesn't have the queue :(

I've now restored the proper monitoring to the server and ran ATTO - first thing is the shocking performance with ATTO which tells me it's the VHDX is the problem as running ATTO on the host on the same disk the VHDX is stored on gives MUCH better results (sorry no screenshots at the mo).

Queue length when running ATTO never reaches 5 so I don't think that's an issue?

Network wise the wires is 1Gbit for that portion and then there's a hop from my wife's laptop to an Asus Black Knight router - usual connection speed is well over 100mbit so I don't think it's that.
Click to expand...
ATTO defaults to a QD of 4. There's a drop-down which goes up to 10.

Sounds like caching of VHDX is disabled. Googling isn't bringing up anything super-obvious, but it sounds like by default all caching is disabled, which is getting pushed all the way down to the controller.
 
Tried some more concurrent downloads and I'm seeing the queue length hit the high 40s :eek: Max writes/sec only 9.9

fsutil fsinfo ntfsinfo d: gives what looks to be correct info for the VHDX so I don't think it's that unfortunately.
 
Sorry, one more obvious thing: under the Array Manager, have you got the cache enabled across all the arrays? Each array should have a big green tick against it under Controller Cache > Modify Caching Settings. Also, I have mine set to 25% Read / 75% Write.

I assume this is set correctly as you have good performance when writing directly to the drive (not to the VHDX).
 
Is this a VM? Or are you just using a VHDX for some esoteric reason?
 
I just ran a test: created a 1GB VHDX, and ran ATTO on both the VHDX mounted drive and the original drive. The performance on the VHDX drive was all over the place, but still very high (there's obviously a lot of caching going on), e.g. it hit 900MB/s at 16kB block size; even at 4kB block size it still managed 129MB/s. This is off a single WD Red 3TB that is connected to a P410 with 512MB BBWC. So the OS is caching, and the P410 is caching. I'm pretty happy with those results.

Try creating a new test VHDX just for the purposes of benchmarking. I wonder if your VHDX is just horribly fragmented.

My ATTO results.

The VHDX:
u58ozlh.png


The underlying physical drive:
Rj3nRhN.png
 
Cache is 60/40 write/read.

I am beginning to think something is seriously wrong with the VHDX. It's attached to my DC as the folder redirection, and also file storage etc. with the OS and everything else on separate disks. I've tried optimising a few times with no effect. I think you're right, wipe and rebuild time.
 
Going to be a bit of a process to avoid downtime (she doesn't like downtime!). Have cleared some space off my NAS and am moving the VHDX in question to it (at a steady 150MB/Sec so definitiely nothing wrong with the underlying hardware - the WD reds in the NAS are what are preventing it moving the data even faster.

Wondering how to setup the replacement - the VHDX contains film, photos and documents - no OS or anything. I'm wondering about using a 64k strip size for the array and 64k block size when formating the new VHDX - problem I've had is if I read 100 websites I get 100 opinions on what to do!
 
I still don't understand why you are using the VHDX? Why not just a partition and be done with it?

I went through a big exercise a couple of months ago to move all my stuff off Storage Spaces back onto straight NTFS, and I'm much happier.
 
I may end up having to do that, but will be persevering for a while. I like the ability to have replication going so I can failover if need be - and it's also a learning curve for work where increasingly I'm needing to know how things work including we'll be moving to a virtualised environment in the not so distant future.
 
You must log in or register to reply here.
Back
Top Bottom