Brendan Gregg

NOVEMBER 29, 2017

Hardware virtualization for cloud computing has come a long way, improving performance using technologies such as VT-x, SR-IOV, VT-d, NVMe, and APICv. At Netflix, we've been using these technologies as they've been made available for instance types in the AWS EC2 cloud. Each dimension has progressed through these stages: 1.

Virtualization 111

Virtualization AWS Hardware Storage 111

Brendan Gregg

FEBRUARY 28, 2023

TCP Extensions for Multipath Operation with Multiple Addresses,” [link] Mar 2020 - [Gregg 20] Brendan Gregg, “Systems Performance: Enterprise and the Cloud, Second Edition,” Addison-Wesley, 2020 - [Hruska 20] Joel Hruska, “Intel Demos PCIe 5.0 Ford, et al., “TCP

Performance 110

Performance Latency Cache Virtualization 110

Brendan Gregg

SEPTEMBER 25, 2021

These strange questions came to the fore back in 2014 when Netflix was switching services from CentOS Linux to Ubuntu, and I helped debug several weird performance issues including one I'll describe here. As a Xen guest, this profile was gathered using perf(1) and the kernel's software cpu-clock soft interrupts, not the hardware NMI.

Speed 126

Speed Java AWS Virtualization 126

The Netflix TechBlog

APRIL 17, 2019

Netflix headquarters circa 2014. What does this example have to do with software development and video encoding? Intel and Netflix announced their collaboration on a software video encoder implementation called SVT-AV1 on April 8, 2019. The reference software typically evolves along with the development of the standard.

Open Source 166

Open Source Scalability Speed Efficiency 166

Brendan Gregg

JULY 4, 2021

It's an exciting time for developments in computer performance, not just for the BPF technology (which I often [write about]) but also for processors with 3D stacking and cloud vendor CPUs (e.g., This was a chance to talk about other things I've been working on, such as the present and future of hardware performance. Ford, et al., “TCP

Performance 52

Performance Latency Hardware Storage 52

All Things Distributed

APRIL 4, 2017

In 2014 and 2015 respectively, AWS opened offices in Stockholm and Espoo, Finland. After finding it cost prohibitive to use colocation centers in local markets where their users are based, iZettle decided to give up hardware. In making the switch to AWS, WOW air has saved between $30,000 and $45,000 on hardware, and software licensing.

AWS 116

AWS Airlines Latency Games 116

ACM Sigarch

APRIL 26, 2018

They introduced the architecture of coarse grain reconfigurable array (CGRA) for statically scheduled data flow computing in HOTCHIPS’17 and its software stack of compiler and linker in ICCAD’17. HPU: Holographic Processing Unit (HPU) is the specific hardware of Microsoft’s Hololens. GPU can also be considered as a special SPU.

Hardware 99

Hardware Artificial Intelligence Architecture Google 99

Brendan Gregg

FEBRUARY 28, 2023

TCP Extensions for Multipath Operation with Multiple Addresses,” [link] , Mar 2020 [Gregg 20] Brendan Gregg, “Systems Performance: Enterprise and the Cloud, Second Edition,” Addison-Wesley, 2020 [Hruska 20] Joel Hruska, “Intel Demos PCIe 5.0 Ford, et al., “TCP

Performance 52

Performance Latency Cache Virtualization 52

Brendan Gregg

NOVEMBER 28, 2017

Hardware virtualization for cloud computing has come a long way, improving performance using technologies such as VT-x, SR-IOV, VT-d, NVMe, and APICv. At Netflix, we've been using these technologies as they've been made available for instance types in the AWS EC2 cloud. Each dimension has progressed through these stages: 1.

Virtualization 40

Virtualization AWS Hardware Storage 40

Brendan Gregg

SEPTEMBER 25, 2021

These strange questions came to the fore back in 2014 when Netflix was switching services from CentOS Linux to Ubuntu, and I helped debug several weird performance issues including one I'll describe here. As a Xen guest, this profile was gathered using perf(1) and the kernel's software cpu-clock soft interrupts, not the hardware NMI.

Speed 52

Speed Java AWS Virtualization 52

Brendan Gregg

SEPTEMBER 25, 2021

These strange questions came to the fore back in 2014 when Netflix was switching services from CentOS Linux to Ubuntu, and I helped debug several weird performance issues including one I'll describe here. As a Xen guest, this profile was gathered using perf(1) and the kernel's software cpu-clock soft interrupts, not the hardware NMI.

Speed 40

Speed Java AWS Virtualization 40

Brendan Gregg

SEPTEMBER 5, 2017

Linux has been adding tracing technologies over the years: kprobes (kernel dynamic tracing), uprobes (user-level dynamic tracing), tracepoints (static tracing), and perf_events (profiling and hardware counters). They use third-party software like Docker or Kubernetes to simplify their administration. People rarely create them manually.

Virtualization 76

Virtualization AWS Engineering Hardware 76

Brendan Gregg

AUGUST 8, 2017

Load averages are an industry-critical metric – my company spends millions auto-scaling cloud instances based on them and other metrics – but on Linux there's some mystery around them. They can be also useful when a single value of demand is desired, such as for a cloud auto scaling rule. I've never seen an explanation.

Latency 111

Latency Metrics C++ Systems 111