Emerging data-centric workloads necessitate systems with highly scalable storage components, not only in terms of capacity, but also in terms of performance and energy efficiency. These requirements, in combination with advances in Storage Class Memory (SCM) technologies, will drive the departure from traditional memory and storage architectures. The focus of this paper is on systems that employ Phase-Change Memory (PCM) to either complement or replace traditional media such, as HDDs, protected DRAM, and Flash. We study possible uses of PCM across the entire stack, and present an analysis of the implications this has for the various components, such as the main memory subsystem, storage subsystems, operating systems, filesystems and databases, as well as of its impact on the overall system architectures. Key problems introduced by the new medium are identified and solutions are outlined, which, we believe, will be very valuable for the designers of next-generation data-centric systems and applications.
Sub-block Wear-leveling for NAND Flash
In NAND Flash memories, the introduction of multi-level cells and the scaling
to smaller cell structures has led to a significant decrease in endurance.
Today, devices built in 20nm technology are typically specified for only 3K
program/erase cycles. Using all cells to their limit is therefore of utmost
importance. Strong error-correction codes are used to extend endurance.
However, existing Flash management schemes retire an entire block when the
first page hits its lifetime limit while other pages in the same block might
still be used. We propose new wear-leveling schemes that can be used in
addition to ECC to substantially extend the lifetime of Flash memories.
Write Amplification Analysis in Flash-Based Solid State Drives
Write amplification is a critical factor limiting the random write performance
and write endurance in storage devices based on NAND-Flash memories such as
solid-state drives (SSD). In this paper, we present a novel probabilistic
model of write amplification for log-structured Flash-based SSDs. The impact
of garbage collection on write amplification is influenced by the level of
over-provisioning and the choice of reclaiming policy: we quantify the impact
of over-provisioning on write amplification analytically and by simulation
assuming workloads of uniformly-distributed random short writes; we propose
modified versions of the greedy garbage-collection reclaiming policy and
compare their performance. Finally, we analytically evaluate the benefits of
separating static and dynamic data in reducing write amplification, and how to
address endurance with proper wear leveling.
Cryptographic Security for a Hish-Speed Distributed File System
Storage systems are increasingly subject to attacks. Cryptographic file
systems mitigate the danger of exposing data by using encryption and integrity
protection methods and guarantee end-to-end security for their clients. This
paper describes a generic design for cryptographic file systems and its
realization in a distributed storage-area network (SAN) file system. Key
management is integrated with the meta-data service of the SAN file system.
The implementation supports file encryption as well as integrity protection
through hash trees. Both techniques have been implemented in the client file
system driver. Benchmarks demonstrate that the overhead is noticeable for
some artificially constructed use cases, but that it is very small for typical
file system applications.
The Potential of Just-In-Time Compilation in Active Networks based on
Network Processors
Byte-code representations in active networks provide architectural neutrality
and code compactness; however, the resulting execution speed is typically poor
due to interpretation overhead. This paper shows that the performance of
capsule-based active networks can benefit from compiling active network
programs into native network processor instructions at traversed routers
(just-in-time compilation). A key aspect of the paper is to demonstrate that
just-in-time compilers for active networks can be fast and small enough for
applicability in the datapath of network processors. The approach has been
implemented based on the SNAP active network framework for the PowerNP network
processor.
Bandwidth Allocation for non-responsive Flows with Active Queue
Management
This paper addresses the problem of configuring active queue management systems
(e.g. WRED and RIO) for service level specifications in Internetworks. In
particular, we focus on Assured Forwarding (AF) for non-responsive flows in
Differentiated Services networks. The difficulty is to determine the correct
queue level thresholds that will result in correct drop rates for various AF
precedence levels under any combination of of-fered loads. A new active queue
management scheme based on control loop is pro-posed that senses not only queue
levels but also rates of queue levels changes and per flow bit rates to
converge automatically to an optimal set of trans-mit fractions. The scheme has
been implemented and tested on a network processor. Results show that the new
active queue management scheme protects assured aggregated flow rates during
periods of congestion. For non-responsive traffic the buffer occupancy level
remains low during 250% offered load.
A Buffer-Management Scheme for Bandwidth and Delay Differentiation using a
Virtual Scheduler
This paper presents a new scalable buffer-management scheme for IP
Differentiated Services. The scheme consists of a Dif-ferentiated Random Drop
(DRD) algorithm using feedback from a virtual scheduler. DRD choses a queue to
perform an early packet drop to avoid congestion according to a specific
probability function. First it will be shown that DRD in conjunction with
first-come first-served scheduling is able to support relative service
differentiation. The virtual scheduler is introduced to enable service
differentiation in terms of bandwidth and delay at the same time. A virtual
sched-uler runs in parallel to the real scheduler and maintains virtual queue
lengths that are being used by the congestion avoidance scheme as a feedback
for packet drop decisions. Scheduling packets for transmis-sion is performed by
the real scheduler only.
Adaptive End-to-End Quality-of-Service Guarantees in IP Networks using an
Active Networking Approach
This paper proposes a framework based on an active networking approach to
efficiently link Quality-of-Service (QoS) descriptions from an application
point of view with an underlying heteroge-neous IP networking
infrastructure. The main goal is to provide building blocks that cooperate to
sense the availability of and deploy distinct QoS capabilities in order to
accomplish adaptive end-to-end service guarantees. The building blocks needed
in a heterogeneous IP network will be introduced and discussed with respect to
safety from abuse of total networking bandwidth, CPU, and memory usage. In
conjunction with a new safety hierarchy and a sandbox environment for
active-code execution, security risks can be bounded to the level of
traditional IP forwarding, control, and management. In particular, the problem
of QoS-parameter translation to provide end-to-end service guarantees is
addressed, and an example using Diffserv, RSVP, and GPRS in a heterogeneous
network is given.
Cryptographic Security for a High-Performance Distributed File System
Storage systems are increasingly subject to attacks. Cryptographic file
systems mitigate the danger of exposing data by using encryption and integrity
protection methods and guarantee end-to-end security for their clients. This
paper describes a generic design for cryptographic file systems and its
realization in a distributed storage-area network (SAN) file system. Key
management is integrated with the meta-data service of the SAN file
system. The implementation supports file encryption and integrity protection
through hash trees. Both techniques have been implemented in the client file
system driver. Benchmarks demonstrate that the overhead is noticeable for some
artificially constructed use cases, but that it is very small for typical file
system applications.
Active Queue Management for Fair Bandwidth Allocation of
Mixed Responsive and Non-Responsive Traffic Using a Closed-Loop
Congestion Control Scheme
Today's known and widely used active queue management (AQM) schemes do not
differentiate between packets from responsive (e.g., TCP sessions) and
non-responsive traffic (e.g., UDP). This results in further widening the gap
of unfair advantage already inherent to non-responsive traffic, as the
responsive sender will significantly reduce its future transmit rate as a
result of the congestion signals. As a simple work-around, responsive and
non-responsive traffic are often assigned distinct AQM parameters. This
approach however requires tuning for each traffic class that potentially
depends on the current or expected offered load. In other words,
responsiveness and TCP-friendliness cannot be estimated easily - not at last
due to short-lived TCP sessions. In this paper we propose a closed-loop
congestion control (CLCC) scheme on top of an existing AQM scheme to achieve
fair bandwidth distribution among concurrent responsive and non-responsive
traffic. The new scheme has the advantage that it does not need to estimate
the level of responsiveness of traffic. We analyze our scheme on top of an
existing rate-based AQM scheme known to approximate max-min fairness, and by
means of simulations show that our extension significantly improves fair
bandwidth allocation for responsive and non-responsive traffic.
PUPRLE: Predictive Active Queue Management Utilizing Congestion Information
Active Queue Management (AQM) tries to find a delicate balance between two
antagonistic Internet queuing requirements: First, buffer space should be
maximized to accommodate the possibly huge transient bursts; second, buffer
occupation should be minimum so as not to introduce unnecessary end-to-end
delays. Traditional AQM mechanisms have been built on heuristics to achieve
this balance, and have mostly done so quite well, but often require manual
tuning or resulted in slow convergence. In contrast, the Purple approach
predicts the impact of its own actions on the behavior of reactive protocols
and thus on the short-term future traffic without per-flow state. Purple
allows much faster convergence of the main AQM parameters, at least towards a
local optimum, thereby smoothing and minimizing both congestion feedback and
queue occupancy. To improve the quality of the prediction, we also passively
monitor (using lightweight operations) information pertaining to the amount of
congestion elsewhere in the network, for example, as seen by flows traversing
this router.
Closed-Loop Congestion Control for Mixed Responsive and Non-Responsive Traffic
Today's known and widely used active queue management (AQM) schemes do not
differentiate between packets from responsive (e.g., TCP sessions) and
non-responsive traffic (e.g., UDP). This results in further widening the gap of
unfair advantage already inherent to non-responsive traffic, as the responsive
sender will significantly reduce its future transmit rate as a result of the
congestion signals. As a simple work-around, responsive and non-responsive
traffic are often assigned distinct AQM parameters. This approach however
requires tuning for each traffic class that potentially depends on the current
or expected offered load. In other words, responsiveness and TCP-friendliness
cannot be estimated easily - not at last due to short-lived TCP sessions. In
this paper we propose a closed-loop congestion control (CLCC) scheme on top of
an existing AQM scheme to achieve fair bandwidth distribution among concurrent
responsive and non-responsive traffic. The new scheme has the advantage that
it does not need to estimate the level of responsiveness of traffic. We
analyze our scheme on top of an existing rate-based AQM scheme known to
approximate max-min fairness, and by means of simulations show that our
extension significantly improves fair bandwidth allocation for responsive and
non-responsive traffic. The simulation results have been verified with a
prototype implementation on the IBM PowerNP 4GS3 network processor.
Creating Advanced Functions on Network Processors: Experience and Perspectives
In this paper, we present five case studies of advanced networking functions
and how a network processor (NP) can provide high-performance and flexible
support for each of them. We first review the basic NP system architectures,
and describe in more detail the IBM PowerNP architecture from a data plane as
well as from a control plane point of view. We introduce models for the
programmer's views of NPs that facilitate a global understanding of NP software
programming. Then, for each case study, we present results from prototypes as
well as general considerations that also apply to a wider range of system
architectures. Namely, we investigate the suitability of NPs for
quality-of-service (active queue management and traffic engineering), header
processing (GPRS tunneling protocol), intelligent forwarding (load-balancing
without flow disruption), payload processing (active networks code
interpretation and just-in-time compilation), and protocol stack termination
(SCTP). Finally, we summarize the key features required by each case study, and
make concluding remarks regarding the future of NPs.
A new Buffer-Management Scheme for IP Differentiated Services
The sophisticated Quality-of-Service (QoS) demands of research, education and
commercial network service providers require new services in current
best-effort Internet architecture. The Internet must enable applications that
demand specific services to profit from a set of differentiated traffic
classes, which support either relative or absolute types of quality of service,
or both. This paper focuses on the design of scalable buffer management and
queueing strategies in a QoS-enabled Internet. A threshold-based buffer
management to be used mainly in core routers is proposed and evaluated. A new
buffer management scheme, called Differentiated Random Drop (DRD) scheme, is
introduced. Combined with simple first-come, first-served (FCFS) scheduling,
the scheme can support differentiated services (Diffserv) that is being
standardized by the IETF.
Adaptive End-to-End Quality-of-Service Guarantees in IP Networks using an
Active Networking Approach
This paper proposes a framework based on an active net-working approach to
efficiently link Quality-of-Service (QoS) descriptions from an application
point of view with an underlying heterogeneous IP networking
infrastructure. The main goal is to sense the availability of and deploy
distinct QoS capabilities in order to accomplish adaptive end-to-end service
guarantees. The building blocks needed in a heterogeneous IP network will be
introduced and dis-cussed with respect to safety in terms of total networking
bandwidth, CPU, and memory usage. In conjunction with a new safety hierarchy
and a sandbox environment for active-code execution, security risks are reduced
to traditional IP forwarding. In particular, the problem of QoS parameter
translation to provide adaptive end-to-end ser-vice guarantees is addressed and
an example using Diffserv and RSVP in a heterogeneous network is given.
The Role of Network Processors in Active Networks
Network processors (NPs) implement a balance between hardware and software
that addresses the demand of performance and programmability in active
networks (AN). We argue that this makes them an important player in the
implementation and deployment of ANs. Besides a general introduction into the
relationship of NPs and ANs, we describe the power of this combination in a
framework for secure and safe capsule-based active code. We also describe the
advantages of offloading AN control point functionality into the NP and how to
execute active code in the data path efficiently. Furthermore, the paper
reports on experiences about implementing active networking concepts on the
IBM PowerNP network processor.
Towards High-performance Active Networking
Network processors have been developed to ease the implementation of new
network protocols in high-speed routers. Being embedded in network interface
cards, they enable extended packet processing at link speed as is required,
for instance, for active network nodes. Active network nodes start using
network processors for extended packet processing close to the link. The
control and configuration of high-performance active network nodes with
network processors such that new services can benefit from the additional
processing capacity offered is nontrivial. In this paper, we present PromethOS
NP which is a modular and flexible router architecture that provides a
framework for dynamic service extension by plugins with integrated support of
network processors, namely the IBM PowerNP 4GS3 network processor. We briefly
introduce the PowerNP architecture in order to show how our active networking
framework maps onto this network processor and provide results from
performance measurements. Owing to architectural similarities of network
processors, we believe that our considerations are also valid for other
network processors.
Buffer Management Scheme for IP Differentiated Services
The demanding quality of service needs of research, education and commercial
network service providers require new services in current best-effort Internet
architecture. The Internet must enable applications that demand specific
services to profit from a set of differentiated traffic classes, which support either
relative or absolute types of quality of service, or both. This diploma thesis
focuses on the design of scalable buffer management and queueing strategies in a
quality-of-service-enabled Internet. A threshold-based buffer management to be
used mainly in core routers is proposed, implemented and tested in the ns
network simulator. A new buffer management scheme, the differentiated random drop
scheme, is introduced. Combined with simple first-come first-served FCFS
scheduling the scheme can support IP differentiated services. The behavior of
the new buffer management is illustrated and compared with other schemes by
means of extensive simulations.
Evaluation of Hierarchical Webcaching and Satellite Distributed Caching
The World Wide Web is growing exponentially and already accounts for a big
percentage of the traffic in the Internet. Often popular Web servers are
overloaded, hot documents travel many times across the same congested links,
and receivers experience slow response times. Cache hit rates can be
significantly increased by having caches cooperate. In this report we
extensively analyze the log entries of the Eurecom Institute and other Squid caches
in order to show what hit rates might be achieved with cooperating caches. We
also discuss how to chose a parent cache out of several sibling caches based
on ping and download round trip times.
Adaptive End-to-End Quality of Service Guarantees in IP Networks
Quality of Service (QoS) in IP networks has made significant advances in the
past decade, which resulted in the standardization of QoS frameworks such as
Integrated Services and Differentiated Services addressing both relative and
absolute service differentiation.
Simultaneously, hardware development has led to new fast and efficient QoS
mechanisms for traffic differentiation and prevention of the dreaded
congestion collapse.
The hardware itself shifted from simple network interface cards to interfaces
built with application-specific integrated circuits, leading to the
introduction of application-specific instruction-set processors, or in other
words, network processors.
This evolution allowed the development of new, more sophisticated algorithms
for QoS support.
However, it can be safely stated today that QoS support still is rarely used.
The increasing gap between the description of QoS parameters and the
capabilities of the underlying hardware, which becomes even more important
when traversing heterogeneous networks consisting of networking equipment from
different manufacturers each having different capabilities, is one of the
reasons for this. The lack of interoperability between QoS mechanisms cannot
be solved solely on the protocol level.
Moreover, QoS requirements differ for each type of application; a
one-size-fits-all solution is not satisfactory, and, depending on the
underlying QoS mechanisms, mapping these requirements to them is
difficult. Therefore QoS deployment is an extremely complex task.
This thesis addresses existing and new QoS mechanisms whose integration,
interaction, and interoperation are not solvable on the protocol level to
build adaptive end-to-end QoS guarantees. To do so, a safe, efficient, and
adaptive framework using active networks (SeaFan) proposed that is
flexible enough to address certain QoS tasks even in the data path. Safety and
security requirements are ensured by the combination of a byte-code language,
the introduction of the resource-bound vector, the definition of a safety
hierarchy, and a sandbox environment.
The minimum set of functionalities in a node model supporting the
active-networking framework is specified. These functionalities are capable of
acting locally on the nodes and globally with respect to the end-to-end
service.
The concept of having several QoS capabilities running at the same time is
explicitly allowed.
New QoS mechanisms are introduced that address relative and absolute
bandwidth differentiation with responsive and non-responsive protocols,
including packet-drop-rate differentiation.
Scalability is ensured by the aggregation of flows and the careful limitation of
the distribution of information, even when acting on the end-to-end service from
within the network.
The excellent performance of the new QoS mechanisms has been shown by means of
simulations in
Created by: roman.pletka
switzerland
org