Demitasse

Andre posted recently about an issue that arises with VMware View Composer pools where a replica, to which a group of desktop VMs are anchored, has a larger than expected storage footprint. 

The issues arises because data has been deleted from inside the master VM, from which the replica VM’s disk image is created by sparsing.  Sparsing is removing any unused blocks to reduce the storage space required. The sparsing process is also used when VMs are converted to thin disks such as by storage VMotion.  The problem is that unused blocks aren’t what is removed, zeroed blocks are removed. To preserve the guest’s view of the disk sparsing leaves intact any block on disk that contains data, even if that data is a deleted file. 

Keeping the replica small is very desirable if there will be many copies on your shared storage, which is what happens without tiered storage configurations.  In Andre’s 2000 seat example a copy of the Replica was stored for roughly every seven desktop VMs, any bloat in the replica was multiplied by about 280 when the View Composer pool was deployed.

The simplest way to resolve the issue is to use sdelete from SysInternals to zero the free space in the master before taking the snapshot that will be the basis for the replica.

Here’s how it looks:

1. My Winodws 7 master, 40GB Hard disk with 26GB of OS and applications installed (yes I currently have everything installed and only thinapp a small selection) produces a similarly sized replica.

2. To create blocks of data from deleted files I copy a few ISO files (8.26GB) to the disk and then delete them.  The OS still reports 26GB consumed but the replica is rather larger, reflecting the non-zeroed blocks previously used by the ISOs.

3. In the VM I run SDelete to zero the free space, this removes the ability to undelete the files.

4. The VM still reports 26GB of consumed disk, however the replica has returned to it’s original size.

The conclusion is that with View Composer pools it’s important to completely clean-up the master image before recomposing your pool, including zeroing free space after you have defragmented the disk.

If you use the View 4.5 feature of tiered storage you can reduce the size multiplication as only a single copy of the replica is required for each pool, however this brings with it the issue of all replica IO going to a single datastore.  Andre has a great article explaining Tiered Storage on View 4.5.

We haven’t managed to line up a guest, so there will be no podcast tonight.

Join us on April 27th to hear about Tintri’s VM aware storage appliance.

If you have an idea for a guest or subject area for the podcast please drop me a line alastair@demitasse.co.nz

Xangati recently released a dedicated dashboard for monitoring VDI deployments, it will be my pleasure to welcome Jagan Jagannathan back to the podcast to discuss how this will help ensure administrator and user happiness through a VDI deployment.

Continue reading →

In my earlier post I outlined how to enable an ESXi server build from the network, now I want to look at the steps required to make the ESXi automatically rebuild when it is network booted. 

Continue reading →

This week I made it onto Andre’s weekly VDI Digest (yay me) and consequently hit a few more people’s horizons.  I’m going to assume that you know who Andre is and already follow his blog and tweets.  One person who I suspect follows Andre is Dwayne Lessner as he commented on my ThinApp user profile size post, leading to me looking at his blog.  I like the content and especially the use of Dilbert, keep it up Dwayne.

VMware View 4.6 with PCoIP Secure Gateway, Enhanced USB device compatibility, Support for Microsoft Windows 7 SP1, iPad Client, ThinApp 4.6.1 and more… A lot has happened in the VMware’s End User Computing space in the last few weeks.

We are pleased to have once again VMware’s APAC Product Marketing Manager for EUC, David Wakeman to talk about the milestones recent delivered by VMware.

Alastair Cooke, that has recently joined VMware Professional Services, is your host for another fantastic episode of the APAC Virtualisation Roundtable.

The broadcast is LIVE via TALKSHOE.com.

You can participate via web accessing http://www.talkshoe.com/tc/75046.

Title: VMware View 4.6 with David Wakeman
Start Time: Wed 16/03/2011 09:00 PM EST (Sydney Time)
Duration (minutes): 45” – 60”
Call ID: 75046

or click the link below

Other Time zones:
EDT (USA) – 7AM
PDT (USA) – 4AM
Perth (Australia) – 7PM
Hong Kong (Hong Kong) – 7PM
Kuala Lumpur (Malaysia) – 7PM
Tokyo (Japan) – 8PM
Auckland (New Zealand) – 11PM
London (UK) – 12 Noon

If you would like to listen pasts episodes click here or download from iTunes.

I can confirm Eric “Scoop” Sloof’s experience, the long awaited View Client for iPad is in the app store.  It brings PCoIP to the iPad and at a much lower price (free) than any competing product.  Make sure you have a View 4.6 environment and have followed Andre’s directions to make it work with PCoIP through a gateway since the iPad client is PCoIP only. 

Also Richard Garsthagen pointed out that there is a dedicated forum in the VMware Communities for the iPad client

The release notes and install guide for View 4.6 spell out the requirements and setup steps for PCoIP through security gateway too.  You should make a habit of reading release notes and other VMware documentation, the answers you seek online are often in the documentation.

I was reading Andre Leibovici’s excellent article on View Floating desktops and comparing it with my feeling about View features and use.  I completely agree that if possible all desktop pools should be floating assignment, ideally using View Composer and refreshing at logoff so users always get the same state when they logon.

One of the key things Andre mentions is controlling the IOPs as storage performance is a top cause of scalability issues with VDI.  My preference for refresh at logoff will cause additional IOPs, this may drive a preference to refresh less frequently.  The other IOPs load is roaming profile load at logon, it was this that I took a look at. 

In my lab I run a Windows 7 floating desktop with refresh at every logoff, so I use roaming profiles and folder redirection for persistence of user state.  I also use ThinApp to get some application into my desktops, especially ones that like to store persistent user data in a non-roaming AppData folder such as FeedDemon.  ThinApp uses a Sandbox folder for each application to store user specific files and registry settings, this can become large particularly with FeedDemon putting it’s cache there.  The default sandbox location is in the user’s roaming AppData folder, consequently bloating the user’s profile and requiring more IOPs at profile load.  In my case my profile size was just under 40MB, of which 33MB was sandbox. 

Luckily it is possible to change the sandbox path in the ThinApp package.  Simply locate the “[BuildOptions]” section of the Package.ini file and either create or modify the “SandboxPath” entry to point to a user specific shared folder.  In my case I have a folder in the user’s home directory and a login script that connects H: drive to the home directory so the section looks like this:

[BuildOptions]
SandboxPath=H:\VDIFolders\Thinstall\

After rebuilding the package and placing it on the deployment share I was able to move the sandbox folder to my home directory.  Now my profile is under 6MB in size, minimising the IOP load at logon and consequently the time required to logon. 

For about the last two years I’ve been trying to find a way to do hands on virtualisation work as well as delivering training.  This week it has finally come to pass, I am now a VMware PSO Consultant as well as being a trainer.  The mix of training and consulting should add value to both roles, the challenge for me will be balancing the commitments.

I’m really excited to be working in consulting and enjoying spending time around the VMware New Zealand team. 

This week I will be talking storage with Otto Le Roux of Datastor, a New Zealand storage distributor with plenty of virtualisation focus.

• SMB Storage
• Unified storage value
• Common design & implementation issues
• iSCSI vs FC vs FCoE.

The broadcast is LIVE via TALKSHOE.com.

You can participate via web accessing http://www.talkshoe.com/tc/75046.

Title: Storage
Start Time: Wed 2/3/2011
   9:00 PM Sydney Time
   11:00 PM Auckland Time
Duration: 45 minutes
Call ID: 75046

or click the link below

If you interested in joining the APAC Virtualization Roundtable initiative to host or guest a episode please reach out to me or one of the other hosts