Unpredictable costs are a growing concern in serverless computing, where applications rely on cloud APIs with complex tiered pricing models. In many deployments, API calls dominate expenses, and a single overlooked design choice can escalate costs by thousands of dollars. Existing tools fall short: provider calculators need unrealistic manual estimates, and dynamic profilers only work post-deployment.

Serverless is an attractive computing model that offers seamless scalability and elasticity; it takes the infrastructure management burden away from users and enables a pay-as-you-use billing model. As a result, serverless is becoming increasingly popular for highly elastic and bursty workloads. However, existing platforms are supported by bloated virtualization stacks, which, combined with bursty and irregular invocations, lead to high memory and latency overheads.

Fine-grained serverless functions power many new applications that benefit from elastic scaling and pay-as-you-use billing model with minimal infrastructure management overhead. To achieve these properties, Function-as-a-Service (FaaS) platforms disaggregate compute and state and, consequently, introduce non-trivial costs due to the loss of data locality when accessing state, complex control plane interactions, and expensive inter-function communication. We revisit the foundations of FaaS and propose a new cloud abstraction, the cloud process, that retains all the benefits of FaaS while significantly reducing the overheads that result from disaggregation. We show how established operating system abstractions can be adapted to provide powerful granular computing on dynamically provisioned cloud resources while building our Process as a Service (PraaS) platform. PraaS improves current FaaS by offering data locality, fast invocations, and efficient communication. PraaS delivers remote invocations up to 17× faster and reduces communication overhead by up to 99%.

Serverless computing allows developers to deploy and scale stateless functions in ephemeral workers easily. As a result, serverless computing has been widely used for many applications, such as computer vision, video processing, and HTML generation. However, we find that the stateless nature of serverless computing wastes many of the important benefits modern language runtimes have to offer. A notable example is the extensive profiling and Just-in-Time (JIT) compilation effort that runtimes implement to achieve acceptable performance of popular high-level languages, such as Java, JavaScript, and Python. Unfortunately, when modern language runtimes are naively adopted in serverless computing, all of these efforts are lost upon worker eviction. Checkpoint-restore methods alleviate the problem by resuming workers from snapshots taken after initialization. However, production-grade language runtimes can take up to thousands of invocations to fully optimize a single function, thus rendering naive checkpoint-restore policies ineffective.

Programming languages offer a number of abstractions such as dynamic typing, sandboxing, and automatic garbage collection which, however, come at a performance cost. Looking back, the most influential programming languages were proposed at a time when Moore’s Law was still in place. Nowadays, post-Moore’s law, scalability, and elasticity become crucial requirements, leading to an increasing tension between programming language design and implementation, and performance. It is now time to discuss the impact of programming languages and language runtimes in the context of scalable and elastic cloud computing platforms with the goal of forecasting their role in the new cloud era.

Function-as-a-Service has emerged as a trending paradigm that provides attractive solutions to execute fine-grained and short-lived workloads referred to as functions. Functions are typically developed in a managed language such as Java and execute atop a language runtime. However, traditional language runtimes such as the HotSpot JVM are designed for peak performance as considerable time is spent profiling and Just-in-Time compiling code. As a consequence, warmup time and memory footprint are impacted. We observe that FaaS workloads, which are short-lived, do not fit this profile.

Function-as-a-Service provides attractive solutions to execute fine-grained and short-lived functions. Functions are typically developed in a managed language and execute atop a language runtime. However, traditional runtimes are designed for peak performance as considerable time is spent profiling and Just-in-Time compiling code. We observe that short-lived FaaS workloads do not fit this profile.

This paper explores automatic heap sizing where developers let the frequency of GC expressed as a target overhead of the application's CPU utilisation, control the size of the heap, as opposed to the other way around. Given enough headroom and spare CPU, a concurrent garbage collector should be able to keep up with the application's allocation rate, and neither the frequency nor duration of GC should impact throughput and latency. Because of the inverse relationship between time spent performing garbage collection and the minimal size of the heap, this enables trading memory for computation and conversely, neutral to an application's performance.

Garbage collection (GC) solutions are widely used in programming languages like Java. Such GC solutions follow prioritized goals and behave differently regarding crucial performance metrics like pause time, throughput, and memory usage. Consequently, running an application using each GC would have an evident impact on the application’s performance, especially on those dealing with massive data and tons of transactions. Nevertheless, it is challenging for a user/developer to pick a GC solution that fits an application’s performance goals. However, surprisingly, there is no tool to help a user/developer on this matter. In this study, we follow an effective methodology to build a heuristic combining throughput and pause time, with diverse heap sizes available, to score four production GCs (G1, Parallel, Shenandoah, and ZGC). Then we propose a system, BestGC, which offers the most suitable GC solution for a user application taking into account the performance goals of the user application.

The serverless computing model leverages high-level languages, such as JavaScript and Java, to raise the level of abstraction for cloud programming. However, today's design of serverless computing platforms based on stateless short-lived functions leads to missed opportunities for modern runtimes to optimize serverless functions through techniques such as JIT compilation and code profiling.

Object Oriented Programming has flourished in many areas ranging from web-oriented microservices, data processing, to databases. However, while representing domain entities as objects is appealing to developers, it leads to data fragmentation, resulting in high memory footprint and poor locality.

The Collections framework is an essential utility in virtually every Java application. It offers a set of fundamental data structures that exploit Java Generics and the Object type in order to enable a high degree of reusability. Upon instantiation, Collections are parametrized by the type they are meant to store. However, at compile-time, due to type erasure, this type gets replaced by Object, forcing the data structures to manipulate references of type Object (the root of the Java type system). In the bytecode, the compiler transparently adds type checking instructions to ensure type safety, and generates bridge methods to enable the polymorphic behavior of parametrized classes. This approach can introduce non-trivial runtime overheads when applications extensively manipulate Collections.

Serverless computing allows users to create short, stateless functions and invoke hundreds of them concurrently to tackle massively parallel workloads. We observe that even though most of the footprint of a serverless function is fixed across its invocations --- language runtime, libraries, and other application state --- today's serverless platforms do not exploit this redundancy. Such an inefficiency has cascading negative impacts: longer startup times, lower throughput, higher latency, and higher cost. To mitigate these problems, we have built Photons, a framework leveraging workload parallelism to co-locate multiple instances of the same function within the same runtime. Concurrent invocations can then share the runtime and application state transparently, without compromising execution safety. Photons reduce function's memory consumption by 25% to 98% per invocation, with no performance degradation compared to today's serverless platforms. We also show that our approach can reduce the overall memory utilization by 30%, and the total number of cold starts by 52%.

Latency sensitive services such as credit-card fraud detection and website targeted advertisement rely on Big Data platforms which run on top of memory managed runtimes, such as the Java Virtual Machine (JVM). These platforms, however, suffer from unpredictable and unacceptably high pause times due to inadequate memory management decisions (e.g., allocating objects with very different lifetimes next to each other, resulting in severe memory fragmentation). This leads to frequent and long application pause times, breaking Service Level Agreements (SLAs). This problem has been previously identified, and results show that current memory management techniques are ill-suited for applications that hold in memory massive amounts of long-lived objects (which is the case for a wide spectrum of Big Data applications).

The cloud is an increasingly popular platform to deploy applications as it lets cloud users to provide resources to their applications as needed. Furthermore, cloud providers are now starting to offer a "pay-as-you-use" model in which users are only charged for the resources that are really used instead of paying for a statically sized instance. This new model allows cloud users to save money, and cloud providers to better utilize their hardware.

Big Data applications suffer from unpredictable and unacceptably high pause times due to Garbage Collection (GC). This is the case in latency-sensitive applications such as on-line credit-card fraud detection, graph-based computing for analysis on social networks, etc. Such pauses compromise latency requirements of the whole application stack and result from applications' aggressive buffering/caching of data, exposing an ill-suited GC design, which assumes that most objects will die young and does not consider that applications hold large amounts of middle-lived data in memory.

Big Data applications suffer from unpredictable and unacceptably high pause times due to bad memory management (Garbage Collection, GC) decisions. This is a problem for all applications but it is even more important for applications with low pause time requirements such as credit-card fraud detection or targeted website advertisement systems, which can easily fail to comply with Service Level Agreements due to long GC cycles (during which the application is stopped). This problem has been previously identified and is related to Big Data applications keeping in memory (for a long period of time, from the GC's perspective) massive amounts of data objects.

Graph data stores are a popular choice for a number of applications: social networks, recommendation systems, authorization and control access, and more. Such data stores typically support both distribution and replication of vertexes across physical nodes.