Solved

Shared LUN/Volume

Posted on 2013-12-11
14
492 Views
Last Modified: 2013-12-13
Say I have 12 VM servers on 3 hosts. They all perform different function like decryption, zip/unzip, etc on large "input" files. To avoid moving these files from one server to another I put them on a file share that all these servers can access, now what used to take 30-40 mins is taking 5-6 hours.

What are my options? someone suggested "Shared LUN" but after reading about that it looks like its frowned upon by the storage experts because exposing a LUN to multiple servers without some coordination leads to your volume going sideways.
0
Comment
Question by:itsmehar
  • 5
  • 4
  • 2
  • +3
14 Comments
 
LVL 117

Assisted Solution

by:Andrew Hancock (VMware vExpert / EE MVE)
Andrew Hancock (VMware vExpert / EE MVE) earned 166 total points
ID: 39712434
You would have to use a Clustered Shared Volume.

Connecting multiple Windows servers to multiple LUNs concurrently will cause corruption.
0
 
LVL 56

Expert Comment

by:Cliff Galiher
ID: 39712441
It'd help to know where the bottleneck is. For example, where is this file share? Is it also a VM? What kind of connectivity does it have? What OS? What disk config?

As a starting point, the network can be a bog bottleneck. Gigabit can get saturated. 10G is a worthy upgrade in these scenarios. Even if you went the shared LUN route, fiber and SAS are faster that gigabit, and iSCSI runs poorly on gigabit. So the speed boost is required regardless.

Then look at the share.  SMB2 will give an immediate bump over SMB1/CIFS. But that requires OS support on both ends. SMB3 would be even better.

Sharing a LUN can be done IF the OS and filesystem supports it. NTFS natively does not, so the advice you read is mostly true. You could, however, set up a cluster with CSVs. I'd personally recommend the SMB3 route, but still.... worth mentioning as an option.
0
 
LVL 19

Expert Comment

by:xterm
ID: 39712442
The fact that they're on a file share is not what's slowing it down as much as the drives in that share are probably maxing out on iops because all the VMs are accessing it simultaneously.

Whatever methodology you come up with for storage, you will need something high performance because of all the I/O on it.  You will find that 7200RPM SATA drives probably won't get it done, and you'll want 10-15K RPM SCSI/SAS drives, lots of spindles, and a RAID 5 architecture rather than a RAID 1.

BTW, a shared LUN isn't a problem generally speaking, but since these are VMs, you can't have dedicated fiber from each into the array.  Furthermore, unless you run some kind of global filesystem (usually SLOW) then each VM won't be able to see anything written by any other VM.  They'd basically each need their own slice of the shared LUN, and it'd be like them each having their own storage.

I guess my question is this - network bandwidth is a lot more expandable and inexpensive than disk storage - why do you feel like you need to avoid moving files around?
0
 
LVL 42

Expert Comment

by:paulsolov
ID: 39712511
Shared Lun will only work with operating systems that support it, typically cluster aware OS.  Sounds like a temp file area, the problem with this is that you're probably getting a lot of latency as multiple sessions are accessing this share.  What we have done is break out a temp file for each server that needs to process this temp data and keep it as a VMDK, you may want to try a decentralized approach and see if performance improves or have multiple share volumes on different spindles.
0
 

Author Comment

by:itsmehar
ID: 39712531
From a bottleneck perspective I have no idea what that is. The obvious things like Bandwidth utilization, CPU/Memory, Disk IO never hit any kind of peek yet the performance degradation is horribly obvious.

The OS is 2008R2 64b, except for the file server where the share resides, which is a VM running 2003 enterprise 64 b OS and the san has 15k SAS drives.

what is happening when you say unzip a file sitting on a remote share, does it bring over the file byte-by-byte and then pushes it back to the share once its done?
0
 
LVL 19

Assisted Solution

by:xterm
xterm earned 167 total points
ID: 39712537
what is happening when you say unzip a file sitting on a remote share, does it bring over the file byte-by-byte and then pushes it back to the share once its done?

Yep!  It has to first read it into memory locally, and then churn it back out across the wire each time.  It's more efficient to do it locally and then push the file back out to where it needs to be, only one copy.
0
 

Author Comment

by:itsmehar
ID: 39712560
"It's more efficient to do it locally and then push the file back out to where it needs to be, only one copy"

hence the shared lun idea. I cannot move a 1 to 25GB file to 10 different servers so they can have access to the file locally, that too 100s of times a day.
0
Netscaler Common Configuration How To guides

If you use NetScaler you will want to see these guides. The NetScaler How To Guides show administrators how to get NetScaler up and configured by providing instructions for common scenarios and some not so common ones.

 
LVL 56

Expert Comment

by:Cliff Galiher
ID: 39712599
That 2003 server is definitely a bottleneck. Running 2008 R2 will show an immediate improvement. 2012/R2 on both ends would be even better.
0
 

Author Comment

by:itsmehar
ID: 39712637
That 2003 server is definitely a bottleneck
could you please elaborate on that a bit, why would that be a bottleneck.
0
 
LVL 56

Accepted Solution

by:
Cliff Galiher earned 167 total points
ID: 39712654
2003 only supports the SMB1 protocol, which is terribly inefficient and decades old. Pushing data back and forth via file shares incurs a heavy overhead cost impacting speed. Vista/2008 introduced the vastly improved SMB2 file sharing protocol. 2012/Win8 introduced SMB3, which is now so efficient, you can run VMs with the VHD stored on a remote file share. So yea, 2003 can definitely be a bottleneck.
0
 

Author Comment

by:itsmehar
ID: 39712731
makes sense, that server is EOL so I can definitely try that.

So if the file server is 2012 (smb3) and the servers accessing it are 2008 (smb2) will that cause any issues? I guess the communication will happen over smb2, the lowest common denominator, is that how it works?
0
 
LVL 56

Expert Comment

by:Cliff Galiher
ID: 39712840
Yep. No problems, you just won't get the benefits of SMB3.
0
 
LVL 55

Expert Comment

by:andyalder
ID: 39713932
You can't share a VHD any easier than you can share a LUN since you still have the problem that VMFS inside the VHD is not a shared filesystem.

By the sounds of it though you are only using one VM at a time to process a huge file/directory and then passing it on to another VM for more processing; if that's the case it may be possible to use batch files to dismount it from one VM and then mount it on the next one in the chain. So long as it is only mounted on one VM at once (or two if one is read-only) then you won't get overwrite corruptions due to the MFT being cached in RAM. You can do this with normal LUNs on a SAN of course but then the LUN presentation has to be managed with the SAN's CLI, it's going to be a lot easier for the controlling script to do it with VHDs since there's just one OS language to use.

There are shared filesystems you can use such as Quantum's StorNext that allow multiple servers and even mixed operating systems to access the same data on a SAN based LUN but they're not cheap.
0
 
LVL 56

Expert Comment

by:Cliff Galiher
ID: 39714491
I don't think anybody mentioned using a VHD to accomplish the goal. I just mentioned that SMB3 is much more network efficient. So much so that Microsoft no supports running Hyper-V VMs where the VHD is stored on a remote SMB3 fileshare. That would have been unheard of just a few years ago, and VHD storage had to be local or use some block-level storage technology (iSCSI/Fibre/SAS for example.) But that in no way implied that a VHD would be shared simultaneously.
0

Featured Post

Find Ransomware Secrets With All-Source Analysis

Ransomware has become a major concern for organizations; its prevalence has grown due to past successes achieved by threat actors. While each ransomware variant is different, we’ve seen some common tactics and trends used among the authors of the malware.

Join & Write a Comment

It Is not possible to enable LLDP in vSwitch(at least is not supported by VMware), so in this article we will enable this, and also go trough how to enabled CDP and how to get this information in vSwitches and also in vDS.
HOW TO: Upload an ISO image to a VMware datastore for use with VMware vSphere Hypervisor 6.5 (ESXi 6.5) using the vSphere Host Client, and checking its MD5 checksum signature is correct.  It's a good idea to compare checksums, because many installat…
Teach the user how to configure vSphere clusters to support the VMware FT feature Open vSphere Web Client: Verify vSphere HA is enabled: Verify netowrking for vMotion and FT Logging is in place or create it: Turn On FT for a virtual machine: Verify …
This Micro Tutorial walks you through using a remote console to access a server and install ESXi 5.1. This example is showing remote access and installation using a Dell server. The hypervisor is the very first component of your virtual infrastructu…

707 members asked questions and received personalized solutions in the past 7 days.

Join the community of 500,000 technology professionals and ask your questions.

Join & Ask a Question

Need Help in Real-Time?

Connect with top rated Experts

17 Experts available now in Live!

Get 1:1 Help Now