© Molten Technologies
By Ivo Murris
This week’s post is a summarised excerpt from a presentation by Ivo Murris at Cloud Expo Europe, 2011. You can see the full presentation here.
To help deliver Desktops as a Service well, one must consider the six major areas of challenge. Given this is a technical review, we will look in detail at the top technical challenges. Besides the technical ones, ensure that other challenges (the expectations of the consumers of the service – the End Users themselves and the licensing for all components of the service and desktops) are also considered before starting.
1. Networking & Remoting – Ensure a client can use their virtual desktops securely in ‘their’ network. Users get the required desktop functionality (e.g. VoIP device support, printing).
Challenges:
• The OS usually has low latency access to the harddisk over the PCI bus. With desktop hosting, multiple desktops are accessing the same storage cluster, often over the network.
• The user interface (UI) is very sensitive to latency.
Some recommendations:
• Separate customer facing networks from storage network and make them redundant.
• Use low latency storage switching gear.
• Minimise UI latency by choosing appropriate remoting protocol and network caching.
• Separate each customer inside their own VLAN and use VRF tags to correctly route traffic.
2. Storage – Ensure that each desktop gets fast access to their hard disk (on central or local server storage).
Challenges:
• Each desktop consumes harddisk space and requires I/O to the storage cluster. Common practices require updating of the OS, anti-virus signature files and applications at fixed times, causing central storage problems.
Some recommendations:
• Don’t use storage geared for server workloads but get dedicated storage for the desktop hosts. High IOPS (75%/25% write/read) and deduplication are required.
• Traditional storage vendors don’t deliver optimal IOPS/$ performance. Look into new vendors delivering SSD-only or hybrid solutions and Cloud characteristics. Ensure performance 24/7, also in degraded state. Assume 15-25 IOPS/desktop.
• Align virtual desktop storage with the hypervisor.
• Optimise the desktop image for virtual delivery by disabling unwanted services and functionality. Measure IOPS/desktop during pilots.
• Think virtual (outside the box): Where appropriate, use dynamic desktop pools with frequently updated golden images, combined with application and user state virtualisation. Use anti-virus applications which work at the level of the desktop host (hypervisor) rather than in each virtual desktop.
3. Servers – Ensure that you get the desktop numbers per rack right while maintaining flexibility.
Challenges:
• Each virtual desktop consumes CPU cycles. Desktop density per server is dependent on resident services and applications on each desktop, similar to thinking about storage. One server ~100 desktops, so reliability is key.
Some recommendations:
• Pilot to gauge virtual desktop density per physical server core. Normal figures are 6 to 10 hosted desktops per physical core, translating to ~96 desktops per dual socket, 12 core server.
• Depending on power and cooling restrictions, density > 8000 desktops per 46U rack is possible (e.g. SGI Rackable, HP Blades).
• From a reliability and networking density perspective, blade technology makes sense.
• It is smarter to build desktop infrastructure Lego style:
o Each rack has its own storage appliances, networking gear, and server nodes.
o Rackbased storage speed is ample to deliver bulk of storage I/O to local desktops.
o Larger customers will require storage replication across these Lego blocks.
o Use Green IT management solutions to turn off desktop nodes or even complete elements when not in use.
4. Offline Access – Management of users requiring offline access. Note this is not necessarily the same as mobile users.
Challenges:
• (Inter)networking has become universal and hosted desktop access should be possible everywhere, also for mobile users. However, in certain cases access might be impossible or very expensive.
Some recommendations:
• Define which corporate applications work offline (no backend requirements).
• Define which user groups need those applications when offline.
• Ensure there’s control of the device (no BYO, real-time interface management, remote destroy).
• Choose a solution that:
o Supports your desktop OS and strategy.
o Is scalable to the required numbers.
o Requires low bandwidth.
o Makes desktop restore quick @ WAN speeds.
What do you think? What have been the top challenges in your experience of virtualisation?
The DaaSler 