Virtualization for Network-based Multicore Telecommunication Systems – Current Status

By Alexander Bachmutsky – Chief Architect at Nokia Siemens Networks

My third post about the virtualization is targeted to provide the current status of the technology. Unfortunately, I would call the status as somewhere between bad to very bad.

Let us analyze a number of potential existing solutions:

• VmWare and KVM are absolutely not applicable for telecommunication systems, because in their current form they cannot be used for real-time networking products, or at least not for their hard and soft real-time parts. It is based on their performance/overhead, processor support and size. Regarding the latter one, refer to my previous posts explaining that a single-point-of-failure critical module cannot be that large.
• Xen open-source hypervisor has been originally targeted for enterprise environment, but latest ARM and PowerPC porting activities (hopefully, MIPS will follow) bring some hope. On the other side, its size is still very large. Also, some architectures might not like the built-in concept of a single master VM (domain zero, or dom0). It is also hard to call it a real-time hypervisor, and the inter-VM communication can become a bottleneck. Some hopes here, but still a long way to go.
• Green Hills Software’s Padded Cell secured hypervisor relies on GHS RTOS, INTEGRITY, to run as a master OS. It supports only x86 and PowerPC architectures and is ported only for INTEGRITY and Linux (Windows and Solaris are also supported on x86 processors, but it is not a common environment for real-time telecommunication systems). The VM separation is based on strict resources partitioning with fixed allocation of CPU, memory and other shared resources between multiple partitions. It also includes tight monitoring of networking activity of every guest OS. One potentially useful feature is the support of Linux binaries allowing Linux applications running in the system without a Linux overhead. It is hard to call it a good generic hypervisor. Implemented in very secured military and avionics applications, not much in the networking infrastructure.
• LynxOS-based virtualization from LynuxWorks supports x86 architecture, and is ported to LynxOS, Linux and Windows. Very similar to GHS’ Padded Cell, including current implemented applications in military and avionics.
• Family of L4 microkernels. Includes L4Ka::Hazelnut, Fiasco, L4Ka::Pistachio, NICTA::L4-embedded, OKL4, still under development Codezero, and others. These compact microkernels (OKL4, for example is less than 10K lines of code) support some levels of virtualization and are implemented on x86, ARM and MIPS processors. The list of paravirtualized OSes includes Linux, SymbianOS, Google’s Android, Microsoft’s Windows, which immediately shows the main focus of L4 family – mobile terminals. Of course, other ports could follow, but it is a limitation today.
• Trango hypervisor had started as a potential solution for telecom products, but it looks like it is also focusing only on the mobile terminals after it was acquired by the VmWare.
• XtratuM open-source low-overhead and small-size hypervisor is developed by the Universidad Politécnica de Valencia in Spain with contributions from the Lanzhou University in China. It is based on fixed cyclic scheduler, and supports x86 and Sparcv8 processors. It has been paravirtualized to Linux and two RTOSes, PartiKle and RTEMS. It was selected by European Space Agency for one of their projects, but its fixed scheduler and limited processors and OSes support is a significant barrier for our intended applications.
• RTS Real-Time hypervisor from Real Time Systems can be very promising solution, because it has hard real-time behavior and fitting architecture, ported to Microsoft Windows, Linux, QNX, VxWorks, On Time RTOS-32, Microware OS-9, and Pharlap ETS operating systems. RTS announced a partnership with Radisys, which could open it for telecommunication infrastructure products, but its support for x86 and Atom processors only is a significant disadvantage. We can only hope that PowerPC, ARM and MIPS processors will be supported in the future, but today its use is still limited.
• Sun Logical Domains is a multithreaded hypervisor with a lot of good features, but limited processors support (x86 and Sparc) and operating systems (Linux and Solaris plus Sun’s simple executive environment) make it also very limited offering.
• VirtualLogix hypervisor has some good potential, supports x86, Atom, ARM, PowerPC and even Texas Instruments’ multicore DSP processors, with operating systems including Linux, Windows, C5, VxWorks, Nucleus, DSP/BIOS, Google’s Android and bare metal proprietary OSes. The concern is whether the company can really maintain so different applications efficiently, and if not, what areas would remain as their target markets. Hopefully, the management overhaul started at the end of 2008 from the appointment of new Executive Chairman of the Board and followed by CEO and CFO in 2009 will stabilize the company. This one still remains to be seen.
• WindRiver hypervisor is very compact and well-known product in many critical applications, it supports Linux and VxWorks, implemented on x86 and PowerPC processors, has high performance, small footprint, determinism, low latency, and high reliability, it can be scaled down to the supervisor role applicable, for example, to the cases with a single VM per core. The biggest concern here is a potential delay in further porting activities because of Intel’s acquisition.

To summarize, it is hard to point to even a single solution that would support many processor architectures and many operating systems. It would be really inappropriate if architects start selecting their hardware and software based on the hypervisor availability. On the other side, selection of unsupported hardware or software might bring your virtualization dreams to a very clear stop sign. Let us all hope that the situation will change soon.

VN:F [1.9.1_1087]

 

http://www.multicorepacketprocessing.com/virtualization-for-network-based-multicore-telecommunication-systems-%E2%80%93-current-status/