Contact Jon.Crowcroft, or Julian Chesterfield at Citrix (ask) Novel storage location, power-aware filesystems Disk/IO throughput performance Network based storage motion Xen-based inter-VM disk IO File-based virtual device drivers Power efficient storage location algorithms Some suggestions of topics a) Based on the principles of File-based object storage for Virtual machine= s, design and develop a system to monitor file access and sort nodes into o= rder of popularity in order to build a fast storage motion manager for larg= e datasets. The final deliverable should focus on one particular workload s= uch as VM booting, or kernel build completion time whilst migrating the und= erlying storage across a high latency data link. Some measurement results t= o demonstrate performance and correctness should also be produced. b) Develop and measure the performance of a Fast fileIO driver for interdom= ain communication over the Xen V4V interface. This will require implementat= ion of a frontend FS interface handler, and a backend FS interface handler = with the definition of a simple message passing protocol in between the two= such as XMLRPC or custom solution. Implementation likely to be on Linux as= a kernel driver (could be implemented in userspace initially). Measurement= would compare performance/memory utilisation against inter-domain network = based solution such as NFS. c) Energy measurement of storage access : may involve a number of project d= efinitions, however themes would focus on quantifying storage energy consum= ption in real systems proportionally to the IO workload being produced. Cha= racterising the differences in energy consumption for different RAID soluti= ons, disk types, data location options on physical drives, SSD storage vs t= raditional hard drive storage. d) Design of an efficient storage caching harddrive: SSD storage is becomin= g increasingly more affordable but still remains likely to be significantly= more costly than traditional hard drive media. SSDs however provide excell= ent IO access speeds, effectively providing persistent RAM-based storage pe= rformance (e.g. battery backed NVRAM). This project would design and implem= ent a 2 tiered storage architecture providing a fast temporary data cache o= n SSD storage, with an asynchronous data copying mechanism to move data ont= o larger, cheaper storage. Such a system would therefore provide an interes= ting trade-off between high capacity cheap large storage disks, with a smal= ler, faster SSD drive to handle incoming IO load. The project would involve= the design and development of an algorithm to identify the best cache popu= lation approach for minimizing realtime dependency on the slower storage, a= nd ensuring data consistency between the 2 devices. The project would produ= ce performance comparison results of the SSD-based caching solution vs trad= itional disk access. Benefits to Student: Both project suggestions a and b will provide detailed exposure to low-leve= l filesystem semantics and Operating system virtualisation. The student wil= l have the opportunity to learn about inter-domain memory management and IO= performance in large distributed systems. The projects have strong relevan= ce to future virtualisation environments and especially cloud workload depl= oyments. Projects c and d have relevance in the understanding of power and performan= ce of storage, particularly in data centres for IO intensive applications. = As the IT industry continues to strive towards more efficient and lower pow= er solutions, such knowledge will be highly relevant.