7/11
KEY NOTE: Surprize-Inspired Networking
- what will be surplus?
- Edge capacity
- High resolution sensors
- how will future apps be different?
- they will be proactive
- E2C (Edge-to-Cloud)
- they will be proactive
[ATC] ZNSwap: un-Block your Swap
- Motivation
- Swap on traditional SSDs
- certain performance anomalies
- Swap on traditional SSDs
- Flash SSDs
- ZNSwap : new sw33ap subsystem for ZNS
[OSDI] BlockFlex
- HarvestVMs: Promising for Improved Resource Utilizaion
- Storage harvesting is challenging
- Insufficient virtualizaition support
- Data security concern
- Storage performance interference
- Software-defined flash
[OSDI] MemLiner2
Observations
App and GC are just temporally unaligned
Changing object access order in GC is possible
2
[OSDI] Carbink: Fault torlerant far Memory
- Memory provisioning is hard, as memory is limited by server physical memory
- Replication vs Erasure Coding(EC)
[OSDI] Demystifying and Checking Silent Semantic Violations in Large Distributed Systems
[OSDI] RESIN
[ATC] Sift
- Bugs on distributed systems
- The power of refinement
- Concise specification
- Scaling to complex systems
- Dealing with undeciability
[OSDI] ListDB
- DRAM+PM Hybrid index/KV-store
- Challenges to avoid write stalls
- DRAM-PM latency gap
- WA
- PM is more sensitive to NUMA effects than DRAM
- NUMA-aware braided skiplist
[OSDI] Odinfs
- Design goal
- Maximize PM performance
- Maintain…
- Key
- Decouple PM access from app threads
- Fixed number of delegation threads limit concurrent access
- Delagation threads always perform localized PM access
- Utilize aggregated bandwidth
[OSDI] Durinn
- Durable linearizability bugs
[ATC] Email Delivery in the 21st Century
[ATC] AddrMiner
7/12
[Keynote] Truthworthy Open Source: The Consequences of Success
- Security problems
- free, but ‘as is’
Curation: Adding a layer of accountability
- Two roles :Distribution vs Accountability
- Distribution: maximize availability of AS IS sw
- Accountablility: supported, more secure versions, perhaps for a fee
- Open source with “Support”
- Focus on fixing vulnerabilites, not just findingtthem
[OSDI] Trinity
- mobile emulator is more than a traditional VM
- Graphics processing is the key in mobile emulator
- Prev work
- RPC
- Async driver commands
- Breaking virtualization
- Goal
- mobile emulator with high efficiency and compatibility
- Graphics projection
- Elasic flow control
- Adaptive data teleporting
[OSDI] ORION
Optimize serverless DAG execution
-
[OSDI] XRP: In-Kernel Storage Functions with eBPF
- Kernel SW is becomeing the bottleneck for storage
- Bypass kernel to eliminate overhead
- Kernel bypass
- pros
- No storage stack overhead
- cons
- pros
- XRP: A framework for in-kernel storage functions
[OSDI] TriCache
-
[ATC] Memory Harvesting in Multi-GPU Systems with Hierarchical Unified Virtual Memory
[OSDI] Verifying the DaisyNFS concurrent and crash-safe filesystem
- Challenges in veritying FS
- Crashes
- Concurrency
- DaisyFS: FS code implemented with transactions
[OSDI] Fast, Automated Inference of Inductive Invariants for Verifying Distributed Protocols
- When accessing app(VM client) data, hypervisor have full previlege.
- But hypervisor is vurnelable
[ATC] IPLFS
[ATC[ Vigil-KV
[ATC] Pacman
- Problem
- High-latency index traversal
- Extra persistence work
- Excessive small random accesses
- Design
- Traverse index with shortcut
- Redesign comapction pipeline
- Separate hot-cold data
- Reduce excessive PM accesses
7/13
[Keynote] The Computer and Information Science and Engineering Landscape: A Look Forward
[OSDI] KSplit
- Driver vulerabilites
- Driver isolation architecture
- Challenge
- Complex data exchange / structures
- Low-level Kernel/C idioms
- Concurrency primitives
- Goals
- Isolate the complete driver
- Identify
- shared/private data
- . . .
[ATC] FlatFS: Flatten Hierarchical FS Namespace on NVMs
- Motivation
- Expensive namespace tree traversal
- Traversing tree causes…
- indirect memory access
- random memory access
- Traversing tree causes…
- Expensive namespace tree traversal
[ATC] StRAID
[ATC] Vinter
- Crash consistency
- goal: recover semantically correct state
댓글