Emmett Witchel
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Emmett Witchel has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
- Zhipeng Jia
Google LLC, Seattle, Washington, United States
5 shared publications - Newton Ni
The University of Texas at Austin, Austin, Texas, United States
3 shared publications - Yan Sun
University of Illinois Urbana-Champaign, Champaign, Illinois, USA
3 shared publications - Zhiting Zhu
The University of Texas at Austin, Austin, Texas, USA and NVIDIA, Santa Clara, California, USA
3 shared publications - Alan M. Dunn
The University of Texas at Austin
2 shared publications - Michael Z. Lee
The University of Texas at Austin
2 shared publications
Latest Publications
- Open Access
- 28 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 211-228
Borges: A Low-Latency Distributed Shared Log on a CXL Memory/SSD Hybrid
Fault-tolerant, distributed shared logs are a useful substrate for distributed applications, but existing designs coordinate shard servers, sequencers, and replicas over the network, so append and replay latency is dominated by message exchange. We present Borges, to our knowledge the first fault-tolerant shared log whose data and ordering state reside in a DRAM/SSD hybrid device accessed with Compute Express Link (CXL). Because current CXL pods lack inter-host cache coherence, Borges assigns every shared metadata word a single writer, atomically publishes each committed prefix as a matching global cut and stream index pair, and tracks each stream with compact byte offsets, so replay stays cheap even when a stream spans shards. We evaluate Borges on Samsung's CMM-H device against CXL ports of Scalog and Boki, using a linearizable lock, a key-value store, and a fault-tolerant workflow with exactly-once semantics. Borges achieves 2.87× lower append p99 latency and 2.16× higher throughput than the Scalog port, and for the Retwis workflow application it lowers p50 latency by 5.7×/22.5× relative to the Scalog/Boki ports.
- 28 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 211-228
Open AccessBorges: A Low-Latency Distributed Shared Log on a CXL Memory/SSD Hybrid
Fault-tolerant, distributed shared logs are a useful substrate for distributed applications, but existing designs coordinate shard servers, sequencers, and replicas over the network, so append and replay latency is dominated by message exchange. We present Borges, to our knowledge the first fault-tolerant shared log whose data and ordering state reside in a DRAM/SSD hybrid device accessed with Compute Express Link (CXL). Because current CXL pods lack inter-host cache coherence, Borges assigns every shared metadata word a single writer, atomically publishes each committed prefix as a matching global cut and stream index pair, and tracks each stream with compact byte offsets, so replay stays cheap even when a stream spans shards. We evaluate Borges on Samsung's CMM-H device against CXL ports of Scalog and Boki, using a linearizable lock, a key-value store, and a fault-tolerant workflow with exactly-once semantics. Borges achieves 2.87× lower append p99 latency and 2.16× higher throughput than the Scalog port, and for the Retwis workflow application it lowers p50 latency by 5.7×/22.5× relative to the Scalog/Boki ports.
- Open Access
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 528-545
C xlalloc : Safe and Efficient Memory Allocation for a CXL Pod
A Compute Express Link (CXL) pod is a group of hosts that share CXL-attached memory. A memory allocator for a CXL pod faces novel challenges: (1) CXL devices may not fully support inter-host hardware cache coherence (HWcc), (2) the allocator may be concurrently accessed from different processes, and (3) with more hosts, failures become more likely. We present cxlalloc, a user-space memory allocator that addresses these challenges through careful metadata layout and new protocols to maintain cache coherence in software, coordinate memory mappings across processes, and recover from crashes. Cxlalloc uses compare-and-swap (CAS) for efficient synchronization; to support CXL devices with no HWcc, we present a memory-based CAS (mCAS) primitive implemented in an FPGA. Experiments with in-memory key-value store workloads demonstrate that cxlalloc retains competitive performance while enabling new use-cases. Experiments with a commercial CXL device show that cxlalloc can achieve 80% of its maximum allocation throughput using mCAS.
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 528-545
Open AccessC xlalloc : Safe and Efficient Memory Allocation for a CXL Pod
A Compute Express Link (CXL) pod is a group of hosts that share CXL-attached memory. A memory allocator for a CXL pod faces novel challenges: (1) CXL devices may not fully support inter-host hardware cache coherence (HWcc), (2) the allocator may be concurrently accessed from different processes, and (3) with more hosts, failures become more likely. We present cxlalloc, a user-space memory allocator that addresses these challenges through careful metadata layout and new protocols to maintain cache coherence in software, coordinate memory mappings across processes, and recover from crashes. Cxlalloc uses compare-and-swap (CAS) for efficient synchronization; to support CXL devices with no HWcc, we present a memory-based CAS (mCAS) primitive implemented in an FPGA. Experiments with in-memory key-value store workloads demonstrate that cxlalloc retains competitive performance while enabling new use-cases. Experiments with a commercial CXL device show that cxlalloc can achieve 80% of its maximum allocation throughput using mCAS.
- Open Access
- 30 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 637-653
Impeller: Stream Processing on Shared Logs
- 30 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 637-653
Open AccessImpeller: Stream Processing on Shared Logs
- Open Access
- 13 November 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Computer Systems
- Vol. 42 (3-4) , 1-35
Boki: Towards Data Consistency and Fault Tolerance with Shared Logs in Stateful Serverless Computing
Bokiis a new serverless runtime that exports a shared log API to serverless functions. Boki shared logs enable stateful serverless applications to manage their state with durability, consistency, and fault tolerance. Boki shared logs achieve high throughput and low latency. The key enabler is the metalog , a novel mechanism that allows Boki to address ordering, consistency and fault tolerance independently. The metalog orders shared log records with high throughput and it provides read consistency while allowing service providers to optimize the write and read path of the shared log in different ways. To demonstrate the value of shared logs for stateful serverless applications, we build Boki support libraries that implement fault-tolerant workflows, durable object storage, and message queues. Our evaluation shows that shared logs can speed up important serverless workloads by up to 4.2×.
- 13 November 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Computer Systems
- Vol. 42 (3-4) , 1-35
Open AccessBoki: Towards Data Consistency and Fault Tolerance with Shared Logs in Stateful Serverless Computing
Bokiis a new serverless runtime that exports a shared log API to serverless functions. Boki shared logs enable stateful serverless applications to manage their state with durability, consistency, and fault tolerance. Boki shared logs achieve high throughput and low latency. The key enabler is the metalog , a novel mechanism that allows Boki to address ordering, consistency and fault tolerance independently. The metalog orders shared log records with high throughput and it provides read consistency while allowing service providers to optimize the write and read path of the shared log in different ways. To demonstrate the value of shared logs for stateful serverless applications, we build Boki support libraries that implement fault-tolerant workflows, durable object storage, and message queues. Our evaluation shows that shared logs can speed up important serverless workloads by up to 4.2×.
- Open Access
- 3 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 41-50
Lupin: Tolerating Partial Failures in a CXL Pod
A compute express link (CXL) pod is a collection of hosts attached to a CXL memory module. It provides an opportunity to port single-host shared-memory programs to execute on multiple hosts in a CXL pod, where the ported application achieves higher performance than a distributed application that uses network for coordination. The cost of performance scaling on a CXL pod is that applications should tolerate partial failures, where one process or operating system fails or reboots. Lupin is system software that includes kernel modifications and user-level libraries to help applications remain available while they recover from partial failures using the contents of CXL memory.
- 3 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 41-50
Open AccessLupin: Tolerating Partial Failures in a CXL Pod
A compute express link (CXL) pod is a collection of hosts attached to a CXL memory module. It provides an opportunity to port single-host shared-memory programs to execute on multiple hosts in a CXL pod, where the ported application achieves higher performance than a distributed application that uses network for coordination. The cost of performance scaling on a CXL pod is that applications should tolerate partial failures, where one process or operating system fails or reboots. Lupin is system software that includes kernel modifications and user-level libraries to help applications remain available while they recover from partial failures using the contents of CXL memory.
- Open Access
- 27 April 2024
- Published by Association for Computing Machinery (ACM)
- p. 2
Challenges and Opportunities for Systems Using CXL Memory
We are at the start of the technology cycle for compute express link (CXL) memory, which is a significant opportunity and challenge for architecture, operating systems, and programming languages. The 3.0 CXL specification allows multiple, physically attached hosts to dynamically share memory. We call such a configuration a CXL pod. Pods provide an intermediate hardware configuration between a network of machines, each with their private memory, and a shared memory multiprocessor with a unified memory, accessible to all machines. This talk will discuss system support for single-node applications to attain scalable performance and high availability across a CXL pod as well as pointing out likely technical challenges for future systems. Along with the technical content, the talk categorizes computer systems research using the model of storytelling with a beginning, a middle and an end. We also examine the fascination popular culture has with personal aha moments and weigh their importance for a group working on an impending submission deadline.
- 27 April 2024
- Published by Association for Computing Machinery (ACM)
- p. 2
Open AccessChallenges and Opportunities for Systems Using CXL Memory
We are at the start of the technology cycle for compute express link (CXL) memory, which is a significant opportunity and challenge for architecture, operating systems, and programming languages. The 3.0 CXL specification allows multiple, physically attached hosts to dynamically share memory. We call such a configuration a CXL pod. Pods provide an intermediate hardware configuration between a network of machines, each with their private memory, and a shared memory multiprocessor with a unified memory, accessible to all machines. This talk will discuss system support for single-node applications to attain scalable performance and high availability across a CXL pod as well as pointing out likely technical challenges for future systems. Along with the technical content, the talk categorizes computer systems research using the model of storytelling with a beginning, a middle and an end. We also examine the fascination popular culture has with personal aha moments and weigh their importance for a group working on an impending submission deadline.
- Top Cited PapersOpen Access
- 17 April 2021
- Published by Association for Computing Machinery (ACM)
- p. 152-166
Nightcore: efficient and scalable serverless computing for latency-sensitive, interactive microservices
The microservice architecture is a popular software engineering approach for building flexible, large-scale online services. Serverless functions, or function as a service (FaaS), provide a simple programming model of stateless functions which are a natural substrate for implementing the stateless RPC handlers of microservices, as an alternative to containerized RPC servers. However, current serverless platforms have millisecond-scale runtime overheads, making them unable to meet the strict sub-millisecond latency targets required by existing interactive microservices. We present Nightcore, a serverless function runtime with microsecond-scale overheads that provides container-based isolation between functions. Nightcore’s design carefully considers various factors having microsecond-scale overheads, including scheduling of function requests, communication primitives, threading models for I/O, and concurrent function executions. Nightcore currently supports serverless functions written in C/C++, Go, Node.js, and Python. Our evaluation shows that when running latency-sensitive interactive microservices, Nightcore achieves 1.36×–2.93× higher throughput and up to 69% reduction in tail latency.
- 17 April 2021
- Published by Association for Computing Machinery (ACM)
- p. 152-166
Top Cited PapersOpen AccessNightcore: efficient and scalable serverless computing for latency-sensitive, interactive microservices
The microservice architecture is a popular software engineering approach for building flexible, large-scale online services. Serverless functions, or function as a service (FaaS), provide a simple programming model of stateless functions which are a natural substrate for implementing the stateless RPC handlers of microservices, as an alternative to containerized RPC servers. However, current serverless platforms have millisecond-scale runtime overheads, making them unable to meet the strict sub-millisecond latency targets required by existing interactive microservices. We present Nightcore, a serverless function runtime with microsecond-scale overheads that provides container-based isolation between functions. Nightcore’s design carefully considers various factors having microsecond-scale overheads, including scheduling of function requests, communication primitives, threading models for I/O, and concurrent function executions. Nightcore currently supports serverless functions written in C/C++, Go, Node.js, and Python. Our evaluation shows that when running latency-sensitive interactive microservices, Nightcore achieves 1.36×–2.93× higher throughput and up to 69% reduction in tail latency.
- Open Access
- 1 May 2013
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 403 (10816011) , 319-333
Anon-Pass: Practical Anonymous Subscriptions
We present the design, security proof, and implementation of an anonymous subscription service. Users register for the service by providing some form of identity, which might or might not be linked to a real-world identity such as a credit card, a web login, or a public key. A user logs on to the system by presenting a credential derived from information received at registration. Each credential allows only a single login in any authentication window, or epoch. Logins are anonymous in the sense that the service cannot distinguish which user is logging in any better than random guessing. This implies unlinkability of a user across different logins. We find that a central tension in an anonymous subscription service is the service provider's desire for a long epoch (to reduce server-side computation) versus users' desire for a short epoch (so they can repeatedly "re-anonymize" their sessions). We balance this tension by having short epochs, but adding an efficient operation for clients who do not need unlinkability to cheaply re-authenticate themselves for the next time period. We measure performance of a research prototype of our protocol that allows an independent service to offer anonymous access to existing services. We implement a music service, an Android-based subway-pass application, and a web proxy, and show that adding anonymity adds minimal client latency and only requires 33 KB of server memory per active user.
- 1 May 2013
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 403 (10816011) , 319-333
Open AccessAnon-Pass: Practical Anonymous Subscriptions
We present the design, security proof, and implementation of an anonymous subscription service. Users register for the service by providing some form of identity, which might or might not be linked to a real-world identity such as a credit card, a web login, or a public key. A user logs on to the system by presenting a credential derived from information received at registration. Each credential allows only a single login in any authentication window, or epoch. Logins are anonymous in the sense that the service cannot distinguish which user is logging in any better than random guessing. This implies unlinkability of a user across different logins. We find that a central tension in an anonymous subscription service is the service provider's desire for a long epoch (to reduce server-side computation) versus users' desire for a short epoch (so they can repeatedly "re-anonymize" their sessions). We balance this tension by having short epochs, but adding an efficient operation for clients who do not need unlinkability to cheaply re-authenticate themselves for the next time period. We measure performance of a research prototype of our protocol that allows an independent service to offer anonymous access to existing services. We implement a music service, an Android-based subway-pass application, and a web proxy, and show that adding anonymity adds minimal client latency and only requires 33 KB of server memory per active user.
- Open Access
- 16 March 2013
- Published by Association for Computing Machinery (ACM)
- Vol. 41 (1) , 265-278
InkTag
InkTag is a virtualization-based architecture that gives strong safety guarantees to high-assurance processes even in the presence of a malicious operating system. InkTag advances the state of the art in untrusted operating systems in both the design of its hypervisor and in the ability to run useful applications without trusting the operating system. We introduce paraverification, a technique that simplifies the InkTag hypervisor by forcing the untrusted operating system to participate in its own verification. Attribute-based access control allows trusted applications to create decentralized access control policies. InkTag is also the first system of its kind to ensure consistency between secure data and metadata, ensuring recoverability in the face of system crashes.
- 16 March 2013
- Published by Association for Computing Machinery (ACM)
- Vol. 41 (1) , 265-278
Open AccessInkTag
InkTag is a virtualization-based architecture that gives strong safety guarantees to high-assurance processes even in the presence of a malicious operating system. InkTag advances the state of the art in untrusted operating systems in both the design of its hypervisor and in the ability to run useful applications without trusting the operating system. We introduce paraverification, a technique that simplifies the InkTag hypervisor by forcing the untrusted operating system to participate in its own verification. Attribute-based access control allows trusted applications to create decentralized access control policies. InkTag is also the first system of its kind to ensure consistency between secure data and metadata, ensuring recoverability in the face of system crashes.
- 1 June 2008
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 182-191
CARVE: A Cognitive Agent for Resource Value Estimation
Recently, industry has begun investigating and moving towards utility computing, where computational resources (processing, memory and I/O) are availably on demand at a market cost. On-demand access to computational resources enables fine-grained resource allocation for web-based applications, e.g., the possibility of provisioning for a minimum workload while allowing the rental of additional resources for unexpected workload changes. However, renting additional resources relies on the ability to quickly and accurately estimate the value of the resource. This paper introduces CARVE: a cognitive agent for resource value estimation. CARVE is a machine-learning based approach that learns to predict the change in system value of having more or less system resources. Using only low-level statistics and with no custom instrumentation of the operating system or middleware, CARVE is able to make informed decisions about the return on investment of physical memory when implemented and evaluated on a partitioned system running a multi-partition, multi-process distributed benchmark. We show that CARVE is competitive with static choices of computing resources over a variety of test workloads and also has the ability to outperform all static configurations.
- 1 June 2008
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 182-191
CARVE: A Cognitive Agent for Resource Value Estimation
Recently, industry has begun investigating and moving towards utility computing, where computational resources (processing, memory and I/O) are availably on demand at a market cost. On-demand access to computational resources enables fine-grained resource allocation for web-based applications, e.g., the possibility of provisioning for a minimum workload while allowing the rental of additional resources for unexpected workload changes. However, renting additional resources relies on the ability to quickly and accurately estimate the value of the resource. This paper introduces CARVE: a cognitive agent for resource value estimation. CARVE is a machine-learning based approach that learns to predict the change in system value of having more or less system resources. Using only low-level statistics and with no custom instrumentation of the operating system or middleware, CARVE is able to make informed decisions about the return on investment of physical memory when implemented and evaluated on a partitioned system running a multi-partition, multi-process distributed benchmark. We show that CARVE is competitive with static choices of computing resources over a variety of test workloads and also has the ability to outperform all static configurations.