Publications
-
Bratterud, Alfred; Dankel, Gisle Mikal Nitter; Farajianzadeh, Mohammadamin; Groves, John; Juan, Chengyi & Suetterlein, Joshua
[Show all 9 contributors for this article]
(2026).
No Atomics, No Problem. Developing a RAG Pipeline for Shared CXL Memory.
IEEE transactions on computers.
ISSN 0018-9340.
75(4),
p. 1291–1308.
doi:
10.1109/tc.2026.3667821.
Full text in Research Archive
Show summary
We share early experiments with software development for shared CXL memory on the H3 Falcon C5022 CXL switch. We describe how the different stages in the development of a commercial RAG pipeline were impacted by the presence of CXL memory. All of Wikipedia is split into 46 million text passages which are embedded into a high-dimensional embedding space and subjected to heavy load in single- and multi-host experiments. We observe significant performance advantages available to DuckDB and Faiss without changing the software; we measure up to 12x latency reduction with real workloads and 40x reduction with synthetic workloads on CXL, compared to the same queries run with demand paging on NVMe. We explain the current challenges of working with two disjoint cache coherency domains and explain how the Fabric-Attached Memory File System (famfs) and famfs producer-consumer queues provide synchronization patterns without atomic operations on current x86 CPUs. We then discuss open challenges remaining for high performance atomics and locking mechanisms in shared memory. Lastly we show how famfs page-level interleaving enables near-linear throughput scaling when a second host serves queries from the same Faiss index on shared fabric-attached memory and how shared memory allocations can be orchestrated by Kubernetes in a full vertical RAG deployment.
-
Hammer, Hugo Lewi; Yazidi, Anis; Bratterud, Alfred; Haugerud, Hårek & Feng, Boning
(2017).
Recovering Request Patterns to a CPU Processor from Observed CPU Consumption Data.
In Maglaras, Leandros (Eds.),
Industrial Networks and Intelligent Systems.
Springer.
ISSN 9783319525686.
p. 14–28.
doi:
10.1007/978-3-319-52569-3_2.
Full text in Research Archive
Show summary
Statistical queuing models are popular to analyze a computer systems ability to process different types requests. A common strategy is to run stress tests by sending artificial requests to the system. The rate and sizes of the requests are varied to investigate the impact on the computer system. A challenge with such an approach is that we do not know if the artificial requests processes are realistic when the system are applied in a real setting. Motivated by this challenge, we develop a method to estimate the properties of the underlying request processes to the computer system when the system is used in a real setting. In particular we look at the problem of recovering the request patterns to a CPU processor. It turns out that this is a challenging statistical estimation problem since we do not observe the request process (rate and size of the requests) to the CPU directly, but only the average CPU usage in disjoint time intervals. In this paper we demonstrate that, quite astonishingly, we are able to recover the properties of the underlying request process (rate and sizes of the requests) by using specially constructed statistics of the observed CPU data and apply a recently developed statistical framework called Approximate Bayesian Computing.
-
Tasoulas, Evangelos; Hammer, Hugo Lewi; Haugerud, Hårek; Yazidi, Anis; Bratterud, Alfred & Feng, Boning
(2017).
The concept of workload delay as a quality-of-service metric for consolidated cloud environments with deadline requirements.
In Gkoulalas-Divanis, Aris (Eds.),
2017 IEEE 16th International Symposium on Network Computing and Applications (NCA).
IEEE (Institute of Electrical and Electronics Engineers).
ISSN 9781538614655.
p. 1–5.
doi:
10.1109/NCA.2017.8171342.
Full text in Research Archive
Show summary
Virtual Machine (VM) consolidation in the cloud has received significant research interest. A large body of approaches for VM consolidation in data centers resort to variants of the bin packing problem which tries to minimize the number of deployed physical machines while meeting the Service-Level-Agreement (SLA) constraints. In this paper we introduce the concept of workload delay as a Quality-of-Service (QoS) metric that captures directly the resulting degradation that a cloud user would experience in the case where the SLA is violated. Our results, that are based on real-life trace-based simulations, show that consolidating VMs based on the level of utilization results in little control over the resulting delay, a particularly significant drawback when running jobs with deadline requirements, while we are able to control the delay much better if we take into account our suggested metric of the delay.
-
Hammer, Hugo Lewi; Yazidi, Anis; Bratterud, Alfred; Haugerud, Hårek & Feng, Boning
(2017).
A Queue Model for Reliable Forecasting of Future CPU Consumption.
Mobile Networks and Applications.
ISSN 1383-469X.
p. 1–14.
doi:
10.1007/s11036-017-0880-3.
Full text in Research Archive
Show summary
Statistical queuing models are popular to analyze a computer systems ability to process different types requests. A common strategy is to run stress tests by sending artificial requests to the system. The rate and sizes of the requests are varied to investigate the impact on the computer system. A challenge with such an approach is that we do not know if the artificial requests processes are realistic when the system is applied in a real setting. Motivated by this challenge, we develop a method to estimate the properties of the underlying request processes to the computer system when the system is used in a real setting. In particular we look at the problem of recovering the request patterns to a CPU processor. It turns out that this is a challenging statistical estimation problem since we do not observe the request process (rate and size of the requests) to the CPU directly, but only the average CPU usage in disjoint time intervals. In this paper we demonstrate that, quite astonishingly, we are able to recover the properties of the underlying request process (rate and sizes of the requests) by using specially constructed statistics of the observed CPU data and apply a recently developed statistical framework called Approximate Bayesian Computing. Further we apply the model to forecast future CPU consumption. Our results show that the model forecast future CPU consumption with less error than both the hidden Markov model (HMM) in (Hammer et al. 2016) and an ARIMA model. Another good property of the queue model is that we can forecast the instantaneous CPU consumption at any time point in the future, while the HMM in (Hammer et al. 2016) and time series models are limited to only forecasting the average CPU consumption in disjoint time intervals.
-
Kvadsheim, Reidar; Haugerud, Hårek; Hammer, Hugo Lewi; Bratterud, Alfred & Habib, Laurence
(2015).
Does Clicker Use Improve Exam Scores? A Controlled Randomized Experiment in a Bachelor-Level Course in Software Engineering.
International Journal of Engineering Education.
ISSN 0949-149X.
31(2),
p. 505–520.
Full text in Research Archive
-
Bratterud, Alfred; Walla, Alf-André; Haugerud, Hårek; Engelstad, Paal E. & Begnum, Kyrre
(2015).
IncludeOS: A minimal, resource efficient unikernel for cloud services.
IEEE International Conference on Cloud Computing Technology and Science (CloudCom).
ISSN 2330-2194.
p. 250–257.
doi:
10.1109/CloudCom.2015.89.
Full text in Research Archive
Show summary
The emergence of cloud computing as a ubiquitous platform for elastically scaling services has generated need and opportunity for new types of operating systems. A service that needs to be both elastic and resource efficient needs A) highly specialized components, and B) to run with minimal resource overhead. Classical general purpose operating systems designed for extensive hardware support are by design far from meeting these requirements. In this paper we present IncludeOS, a single tasking library operating system for cloud services, written from scratch in C++. Key features include: extremely small disk-and memory footprint, efficient asynchronous I/O, OS-library where only what your service needs gets included, and only one device driver by default (virtio). As a test case a bootable disk image consisting of a simple DNS server with OS included is shown to require only 158 kb of disk space and to require 5-20% less CPU-time, depending on hardware, compared to the same binary running on Linux.
-
Hammer, Hugo Lewi; Bratterud, Alfred & Fagernes, Siri
(2013).
Crawling JavaScript websites using WebKit - with application to analysis of hate speech in online discussions.
NIKT: Norsk IKT-konferanse for forskning og utdanning.
ISSN 1892-0713.
p. 25–36.
Full text in Research Archive
-
Bratterud, Alfred & Haugerud, Hårek
(2013).
Maximizing hypervisor scalability using minimal virtual machines.
In Society, IEEE (Eds.),
2013 IEEE International Conference on Cloud Computing Technology and Science.
IEEE (Institute of Electrical and Electronics Engineers).
ISSN 9780769550954.
p. 218–223.
doi:
10.1109/CloudCom.2013.11.
Full text in Research Archive
Show summary
The smallest instance offered by Amazon EC2 comes with 615MB memory and a 7.9GB disk image. While small by today's standards, embedded web servers with memory footprints well under 100kB, indicate that there is much to be saved. In this work we investigate how large VM-populations the open Stack hyper visor can be made to sustain, by tuning it for scalability and minimizing virtual machine images. Request-driven Qemu images of 512 byte are written in assembly, and more than 110 000 such instances are successfully booted on a 48 core host, before memory is exhausted. Other factors are shown to dramatically improve scalability, to the point where 10 000 virtual machines consume no more than 2.06% of the hyper visor CPU.
View all works in NVA
Published
Oct. 18, 2024 2:12 PM
- Last modified
Nov. 18, 2024 9:44 PM
