EuroSys 2010 Paris April 13-16, 2010. Overview 7 workshops, 5 tutorials 10 sessions Storage Systems Transactional Memory Real-Time Systems Systems Management.

EuroSys 2010

ParisApril 13-16, 2010

Overview

• 7 workshops, 5 tutorials• 10 sessions

Storage SystemsTransactional MemoryReal-Time SystemsSystems ManagementScheduling

KernelCloudsSecurityBugs and ProfilingMulticast

The Doctoral Workshop

• 2 keynotes• 5 / 15 min. presentations

The Gauleiter

The Don

vs.

Day 1 – Summary

SSD cache compression

Differential RAID

AMD ASF

Legacy Feedback Scheduling ++

Java for UAVs

Expected cost of operator mistakes

Web app testing proxy

SynQuake

Datacenter fingerprints

Day 1 – Best Paper

„Fingerprinting the Datacenter: Automated Classification of Performance Crises“

P. Bodik (Berkeley), M. Goldszmidt (MSR), A. Fox (Berkeley), D.B. Woodard (Cornell), H. Andersen (Microsoft)

Day 1 – Best Paper„Fingerprinting the Datacenter: Automated Classification of Performance Crises“

Day 1 – Best Paper„Fingerprinting the Datacenter: Automated Classification of Performance Crises“

• Evaluation on production cluster at Microsoft (400 machines)

• 77% accuracy for crisis identification, 82% for identification of unseen crises

• On average after 10 minutes

Day 1 – Runner-up

„Differential RAID: Rethinking RAID for SSD Reliability“

M. Balakrishnan (MSR), A. Kadav (U. Wisconsin), V. Prabhakaran, D. Malkhi (MSR)

Day 1 – Runner-up„Differential RAID: Rethinking RAID for SSD Reliability“

Correlated failure probability increases

Day 1 – Runner-up„Differential RAID: Rethinking RAID for SSD Reliability“

Distribution provably converges

Day 1 – Runner-up„Differential RAID: Rethinking RAID for SSD Reliability“

• Simulation using BER data for 12 chips

• Implementation for performance

Day 2 – Summary

Bias scheduling (x2)

Energy-efficient scheduling

Yet-another-hypervisor™

NIC Driver Reverse Engineer-ator

Surviving kernel crashes

Residue objects

Map-Reduce to SQL converter

Return-less kernels

Q-states for clouds

Delay scheduling on clusters

Declarative cloud programmingWeakly consistentaccess control

Day 2 – Best Paper†

„Otherworld - Giving Applications a Chance to Survive OS Kernel Crashes“

Alex Depoutovitch (U. Toronto), Michael Stumm (U. Toronto)

† according to the conference audience

Day 2 – Best Paper„Otherworld - Giving Applications a Chance to Survive OS Kernel Crashes“

Crash

Launch crash kernel 1) Restore processes2) Migrate resources3) Run crash procedures

Day 2 – Best Paper„Otherworld - Giving Applications a Chance to Survive OS Kernel Crashes“

• Fault injection – 4,000 experiments

• 97-98% recovery success rate

„We have had a similar facility in Minix for years. However, we found that you need to do a lot more experiments as there are some really weird and rare issues that can occur [...] You want to do millions of tests, not just a few thousand.“

– Andrew Tanenbaum [paraphrased from memory]

Day 2 – Best Paper„Otherworld - Giving Applications a Chance to Survive OS Kernel Crashes“

• Open questions- What is re-inited and what is restored? (e.g. daemons)

- What about e.g. disk access during crash and resulting corruption?

- Downtime due to microreboot?

- Modification effort and specificity

Day 2 – Craziest Papers†

„Reverse Engineering of Binary Device Drivers with RevNIC“

Vitaly Chipounov (EPFL), George Candea (EPFL)

† as judged by the presenter...

„Defeating Return-Oriented Rootkits With Return-less Kernels“

J. Li, Z. Wang, X. Jiang, M. Grace, S. Bahram (North Carolina State University)

Day 2 – Craziest Papers„Reverse Engineering of Binary Device Drivers with RevNIC“

Day 2 – Craziest Papers„Reverse Engineering of Binary Device Drivers with RevNIC“

Day 3 – Summary

Ummmm.....

Detect and prevent atomicity violations

Automated software debugging

Cache profiling

Dr. Multicast

Next 700 BFT protocols