Posts

Off to EuroSys 2010 in Paris

Presented our paper on compressed SSD caching in the ACM/SIGOPS European Conference on Computer Systems (EuroSys 2010) in Paris.

Here are the presentation slides (PDF).

How to count Lines of Code (LOC) using grep and wc

Contrary to the CLOC tool (http://cloc.sourceforge.net) I argue that comments in the code are useful and should be counted in the “real” lines of code. I think that this gives a nice incentive to developers for commenting their code. Of course that holds assuming that you are not trying to fool anyone by artificially increasing the lines of your code. If your are trying to fool people, you can still do it with comments being removed (but I’m not going to tell you how).

Parallel Postmark Benchmark

Renaud Lachaize and I have modified the well-known Postmark v1.5 filesystem benchmark to run on a cluster using MPI, in order to measure cluster/parallel filesystem performance. Currently it has been successfully used to measure PVFS2, GFS and Lustre. For experimental results with parallel postmark please check out my publications or the paper “A Comparative Experimental Study of Parallel File Systems for Large-Scale Data Processing” by Zoe Sebepou, Kostas Magoutis, Manolis Marazakis, and Angelos Bilas, presented at the USENIX LASCO ‘08 workshop.

Foxy Web Cache

The Foxy Web Cache is a high-performance web proxy prototype that we used to measure our storage I/O algorithms for web proxies.For the details of storage I/O algorithms and experimental results please check out my publications. Foxy has been implemented and tested on Sun Solaris (2.7-2.8) but should run unmodified on other Unices like Linux, since it uses standard Unix library calls. In our publications Foxy outperforms the Squid web cache by a factor of 4 in throughput (www requests served per second), while maintaining the same low response time as Squid.

The Network RamDisk (NRD)

The Network RamDisk is a device that uses main memory of remote workstations in a workstation cluster (interconnected by a high performance network) as a faster-than-disk storage device. It has been implemented both on the Linux 2.0.x and the Digital Unix 4.0 operating systems, as a block device driver without any modifications to the kernel code. A novel reliability policy (Adaptive Parity Caching) has been designed and implemented along with data replication, making the device tolerant to single workstation crashes.