The Netflix TechBlog

NOVEMBER 9, 2022

At Netflix, we periodically reevaluate our workloads to optimize utilization of available capacity. A quick canary test was free of errors and showed lower latency, which is expected given that our standard canary setup routes an equal amount of traffic to both the baseline running on 4xl and the canary on 12xl. let’s call it GS2?—?to

Hardware 363

Hardware Cache Performance Latency 363

The Morning Paper

JUNE 13, 2019

As we saw with the SOAP paper last time out, even with a fixed model variant and hardware there are a lot of different ways to map a training workload over the available hardware. The following figure highlights how just one of these variables, batch size, impacts throughput and latency on ResNet50.

Hardware 58

Hardware Latency Serverless Energy 58

The Morning Paper

JANUARY 30, 2020

Edge servers are the middle ground – more compute power than a mobile device, but with latency of just a few ms. The current system assumes an application specific regression model is available on the servers which can predict processing time given the current parameters of the job (e.g. in the cloud).

Mobile 104

Mobile Cloud Latency Games 104

Percona

SEPTEMBER 1, 2023

This results in expedited query execution, reduced resource utilization, and more efficient exploitation of the available hardware resources. This reduction in latency ensures that applications and websites provide a more rapid and responsive user experience. Regularly optimizing query structures is vital.

Tuning 52

Tuning Database Performance Hardware 52

Percona

MAY 2, 2023

Since your dataset is 100 GB and you have 500 GB of RAM, you can allocate a significant portion of the available memory to the InnoDB buffer pool. A commonly recommended value for the buffer pool size is between 50% to 75% of the available memory. 16) and monitoring the server’s performance. The answer could be better.

Social Media 118

Social Media Database Servers Cache 118

Dynatrace

JUNE 25, 2020

Dynatrace Managed is intrinsically highly available as it stores three copies of all events, user sessions, and metrics across its cluster nodes. The network latency between cluster nodes should be around 10 ms or less. Turnkey high availability across globally distributed data centers. Dynatrace news.

Availability 246

Availability Hardware Latency Traffic 246

Brendan Gregg

AUGUST 18, 2019

For example, iostat(1), or a monitoring agent, may tell you your average disk latency, but not the distribution of this latency. This example instrumented one of many thousands of available events. For smaller environments, it can be of more use helping eliminate latency outliers. Block I/O latency as a histogram.

Latency 68

Latency C++ Cache Programming 68

The Morning Paper

MAY 14, 2019

Last time around we looked at the DeathStarBench suite of microservices-based benchmark applications and learned that microservices systems can be especially latency sensitive, and that hotspots can propagate through a microservices architecture in interesting ways. on end-to-end latency) and less than 0.15% on throughput.

Big Data 45

Big Data Cloud Performance Hardware 45

Smashing Magazine

AUGUST 3, 2021

Lots can go wrong: a network request fails, a third-party library breaks, a JavaScript feature is unsupported (assuming JavaScript is even available), a CDN goes down, a user behaves unexpectedly (they double-click a submit button), the list goes on. On design systems, CSS/JS and UX. JavaScript is less forgiving. More after jump!

Network 145

Network Engineering Availability Monitoring 145

HammerDB

OCTOBER 23, 2023

HammerDB has always used stored procedures as a design decision because the original benchmark was implemented as close as possible to the example workload in the TPC-C specification that uses stored procedures. Use the performance metrics available in the database first before looking at data further down in the stack.

Benchmarking 62

Benchmarking Database Network C++ 62

Testsigma

JANUARY 30, 2021

The applications are designed using the same code base as Desktop like HTML, Javascript, and CSS. Unlike Native Applications, Mobile web applications do not have to be designed separately for IOS and Android. You just need to hit the URL and launch the application on the available browser on your phone.

Mobile 52

Mobile Testing Website Internet 52

Dotcom-Montior

DECEMBER 8, 2021

ITSM defines how teams design, create, and deliver their services. Software services still require physical devices and hardware for them to function. This helps to improve efficiency and ensures that information is consistent, up-to-date, and available. It is much more than just IT support. Asset Management.

Social Media 90

Social Media Monitoring Latency DevOps 90

Scalegrid

FEBRUARY 8, 2024

Key Takeaways Distributed storage systems benefit organizations by enhancing data availability, fault tolerance, and system scalability, leading to cost savings from reduced hardware needs, energy consumption, and personnel. Variations within these storage systems are called distributed file systems.

Storage 130

Storage Systems Big Data Azure 130

Abhishek Tiwari

JULY 12, 2018

Data Pipeline A data pipeline is a software that ingests data from multiple sources, transforms it and finally makes it available to internal or external products. A data pipeline is a software which runs on hardware. The software is error-prone and hardware failures are inevitable.

Latency 63

Latency Analytics Scalability Engineering 63

All Things Distributed

JULY 13, 2010

Cluster Computer Instances for Amazon EC2 are a new instance type specifically designed for High Performance Computing applications. Today, I am very proud to be a part of the Amazon Web Services team as we truly make HPC available as an on-demand commodity for every developer to use. HPC and Amazon EC2. until today.

Cloud 119

Cloud AWS Automotive Latency 119

Adrian Cockcroft

MARCH 17, 2020

Another problem is that a design control, intended to mitigate a failure mode, may not work as intended. STPA is based on a functional control diagram of the system, and the safety constraints and requirements for each component in the design. Hence, one way to reduce risk is to make systems more observable.

Healthcare 52

Healthcare Government Airlines Systems 52

HammerDB

DECEMBER 5, 2020

A full understanding of why this is important requires some knowledge of the evolution of database hardware and software. The HammerDB TPROC-C workload by design intended as CPU and memory intensive workload derived from TPC-C – so that we get to benchmark at maximum CPU performance at a much smaller database footprint.

C++ 40

C++ Benchmarking Virtualization Open Source 40

Percona

APRIL 17, 2023

Note that the intent of tuning the settings is not exclusively about improving performance but also enhancing the high availability and resilience of the MongoDB database. There is an issue with this, which causes the OS to swap even with memory available. The CFQ works well for many general use cases but lacks latency guarantees.

Best Practices 52

Best Practices Design Tuning Database 52

The Morning Paper

MAY 12, 2019

An open-source benchmark suite for microservices and their hardware-software implications for cloud & edge systems Gan et al., It’s a pretty impressive effort to pull together and make available in open source (not yet available as I write this) such a suite, and I’m sure explains much of the long list of 24 authors on this paper.

Open Source 61

Open Source Hardware Benchmarking Systems 61

Highly Scalable

AUGUST 20, 2013

This system has been designed to supplement and succeed the existing Hadoop-based system that had too high latency of data processing and too high maintenance costs. The design of the in-stream processing engine itself was driven by the following requirements: SQL-like functionality. Fault-tolerance.

Big Data 154

Big Data Processing Lambda Database 154

The Morning Paper

NOVEMBER 10, 2019

It’s been clear for a while that software designed explicitly for the data center environment will increasingly want/need to make different design trade-offs to e.g. general-purpose systems software that you might install on your own machines. The desire for CPU efficiency and lower latencies is easy to understand.

Network 92

Network Transportation Latency Entertainment 92

Alex Russell

APRIL 29, 2021

This post mines publicly available data on the pace of compatibility fixes and feature additions to assess the claim. The information it captures is, however, available going back somewhat further, providing a fuller picture of the trend lines of engine completeness. Count of APIs available from JavaScript by Web Confluence.

Media 145

Media Games Education Engineering 145

All Things Distributed

MARCH 11, 2016

But we couldn’t adopt the old style approach of upgrading systems through a maintenance outage, as many businesses around the world are relying on our platform for 24/7 availability. This is a given, whether you are using the highest quality hardware or lowest cost components. Systems need to keep running even if the “house is on fire.”

AWS 147

AWS Hardware Retail Virtualization 147

The Morning Paper

MARCH 3, 2020

It takes you through the thinking processes and engineering practices behind the design of a key part of the control plane for AWS Elastic Block Storage (EBS): the Physalia database that stores configuration information. EBS uses chain replication for availability, with data flowing from client to primary to replica. NSDI’20.

Database 106

Database AWS Network Design 106

SQL Performance

APRIL 27, 2020

Gen 5 is the primary hardware option now for most regions since Gen 4 is aging out. Microsoft took that idea and built a new compute tier called Azure SQL Database serverless, which became generally available in November 2019. New Hardware Configuration for Provisioned Compute Tier. GB per vCore. Conclusion.

Azure 55

Azure Database Serverless Hardware 55

All Things Distributed

JULY 14, 2015

In traditional database architectures, database engines often run a small search engine or data warehouse engines on the same hardware as the database. DynamoDB Streams simplifies and improves this design pattern with a distributed systems approach. DynamoDB Cross-region Replication.

Database 167

Database Lambda AWS IoT 167

SQL Server According to Bob

SEPTEMBER 26, 2016

When we released Always On Availability Groups in SQL Server 2012 as a new and powerful way to achieve high availability, hardware environments included NUMA machines with low-end multi-core processors and SATA and SAN drives for storage (some SSDs). As we moved towards SQL Server 2014, the pace of hardware accelerated.

Availability 40

Availability Servers Hardware Benchmarking 40

The Morning Paper

MAY 19, 2019

Last week we learned about the [increased tail-latency sensitivity of microservices based applications with high RPC fan-outs. Seer uses estimates of queue depths to mitigate latency spikes on the order of 10-100ms, in conjunction with a cluster manager. So what we have here is a glimpse of the limits for low-latency RPCs under load.

Latency 80

Latency Hardware Network Architecture 80

All Things Distributed

JULY 20, 2015

A cluster is just a pool of compute resources available to a customer’s applications. The agent is written in Go, has a minimal footprint, and is available on GitHub under an Apache license. This architecture affords Amazon ECS high availability, low latency, and high throughput because the data store is never pessimistically locked.

Latency 166

Latency Architecture AWS Open Source 166

John McCalpin

AUGUST 1, 2018

According to Dr. Bandwidth, performance analysis has two recurring themes: How fast should this code (or “simple” variations on this code) run on this hardware? The user environment defines the mapping of MPI ranks to hardware resources (cores, sockets, nodes). The source code to the library may not be available.

Hardware 40

Hardware Transportation Performance Latency 40

The Morning Paper

MARCH 17, 2020

One thing they all had in common is an evaluation using some kind of simulation of the expected behaviour of NVDIMMs, because the real thing wasn’t yet available. The Optane DIMM is the first scalable, commercially available NVDIMM. The read latency for Optane is 2x-3x higher than DRAM. But now it is!

Scalability 59

Scalability Latency Cache Media 59

Percona

JULY 1, 2023

Understanding DBaaS DBaaS cloud services allow users to use databases without configuring physical hardware and infrastructure or installing software. These may be performance, high availability, operational cost, management, capacity planning, scalability, security, monitoring, etc. What is Amazon RDS?

AWS 52

AWS Database Serverless Storage 52

All Things Distributed

JANUARY 18, 2012

a Fast and Scalable NoSQL Database Service Designed for Internet Scale Applications. Today is a very exciting day as we release Amazon DynamoDB , a fast, highly reliable and cost-effective NoSQL database service designed for internet scale applications. Amazon DynamoDB offers low, predictable latencies at any scale. Comments ().

Scalability 167

Scalability Database Ecommerce Latency 167

Dynatrace

OCTOBER 1, 2021

In these modern environments, every hardware, software, and cloud infrastructure component and every container, open-source tool, and microservice generates records of every activity. The architects and developers who create the software must design it to be observed. Benefits of observability.

Metrics 363

Metrics Open Source Monitoring Infrastructure 363

The Morning Paper

JULY 4, 2019

This is a fascinating paper from members of Netflix’s Resilience Engineering team describing their chaos engineering initiatives: automated controlled experiments designed to verify hypotheses about how the system should behave under gray failure conditions, and to probe for and flush out any weaknesses. Safeguards.

Latency 77

Latency Engineering Metrics Traffic 77

The Morning Paper

MAY 21, 2019

This boils down to a single digit µs latency toleration in the tail for far memory, and in addition to security and privacy concerns, rules out remote memory solutions. Zswap is readily available and runs as a swap device in the Linux kernel. Google use zswap to implement their far memory tier.

Software 78

Software Software Google Hardware 78

Smashing Magazine

AUGUST 22, 2021

Because we are dealing with network protocols here, we will mainly look at network aspects, of which two are most important: latency and bandwidth. Latency can be roughly defined as the time it takes to send a packet from point A (say, the client) to point B (the server). Two-way latency is often called round-trip time (RTT).

Performance 105

Performance Network Latency Servers 105