When Tesla CEO Elon Musk first announced the Model 3 over two years ago, he billed it as an “affordable” mainstream electric vehicle that would be available at a price of $35,000 before federal tax credits. More than a year after customers first starting taking deliveries of the Model 3, the mythical $35,000 Standard-Range model still isn’t
In just one more day, AMD’s recently announced Radeon VII graphics card will be available to purchase. That’s also when reviews will hit the web (ours included). In the meantime, several of AMD’s hardware partners have erected Radeon VII product pages on their websites, including both MSI and Gigabyte.
Have you ever sent a message that you instantly regretted? Facebook Messenger now allows you to unsend your typo-ridden, impulsive messages. The catch? You will only have ten minutes to remove the message.
Users will need to open the Facebook Messenger app on their desktop, laptop, or smart device and open the conversation with the desired
Electronic Arts may very well have a hit on its hands with Apex Legends, the free-to-play battle royale shooter developed by Respawn Entertainment, but any financial success it generates will have to wait for another day. As it pertains to EA’s third fiscal quarter ended December 31, 2018, the numbers are down from what the publisher was expecting,
In the ongoing quest to eliminate bezels on smartphones in favor of full-front displays, smartphone makers have gotten creative with their designs. Apple, for example, opted for a bigger version of the notch that was first introduced by Andy Rubin’s Essential Phone, while others are trending towards punch-hole designs—circular cutouts to house
AMD first introduced its 14nm Zen processor architecture in 2017, and it was a turning point for the company’s fortunes in the PC market. This all-new microarchitecture gave rise to Ryzen for the mainstream PC market, Ryzen Threadripper for the high-end desktop (HEDT) market, and EPYC for enterprise customers.
The company followed up on those
Chrome is the most popular browser on the market even though it is far from perfect. One of the biggest complaints that Chrome users have is that the browser tends to be very resource intensive using up memory and excess processing cycles that can slow down the entire system.
It appears that Google is finally ready to do something about
Take-Two has been basking in the glowing press and tremendous sales performance of Red Dead Redemption 2. The game was so well received right out of the gates that it made $725 million in its first three days of availability. Take-Two is the parent company to Rockstar, the developers behind the game. Take-Two has announced its Q3 earnings,
Linuz Henze, a well-known security researcher, has found a particularly troublesome bug in Apple’s macOS. The exploit involves the operating system’s Keychain, which stores passwords, private keys, and other user credentials.
Henze has demonstrated a working app that could be maliciously used to pilfer your Mac’s Keychain, robbing you of
Cluster Management systems such as Mesos, Google Borg, Kubernetes etc. in a cloud scale datacenter environment (also termed as Datacenter-as-a-Computer or Warehouse-Scale Computing – WSC) typically manage application workloads by performing tasks such as tracking machine live-ness, starting, monitoring, terminating workloads and more importantly using a Cluster Scheduler to decide on workload placements.
A Cluster Scheduler essentially performs the scheduling of workloads to compute resources – combining the global placement of work across the WSC environment makes the “warehouse-scale computer” more efficient, increases utilization, and saves energy. Cluster Scheduler examples are Google Borg, Kubernetes, Firmament, Mesos, Tarcil, Quasar, Quincy, Swarm, YARN, Nomad, Sparrow, Apollo etc.
In this blog post, we briefly describe the novel Firmament flow network graph based scheduling approach (OSDI paper) in Kubernetes. We specifically describe the Firmament Scheduler and how it integrates with the Kubernetes cluster manager using Poseidon as the integration glue. We have seen extremely impressive scheduling throughput performance benchmarking numbers with this novel scheduling approach. Originally, Firmament Scheduler was conceptualized, designed and implemented by University of Cambridge researchers, Malte Schwarzkopf & Ionel Gog.
Poseidon-Firmament Scheduler – How It Works
At a very high level, Poseidon-Firmament scheduler augments the current Kubernetes scheduling capabilities by incorporating novel flow network graph based scheduling capabilities alongside the default Kubernetes Scheduler. It models the scheduling problem as a constraint-based optimization over a flow network graph – by reducing scheduling to a min-cost max-flow optimization problem. Due to the inherent rescheduling capabilities, the new scheduler enables a globally optimal scheduling environment that constantly keeps refining the workloads placements dynamically.
Workloads (pods) are bulk scheduled to enable scheduling decisions at massive scale.
Based on the extensive performance test results, Poseidon-Firmament scales much better than Kubernetes default scheduler as the number of nodes increase in a cluster. This is due to the fact that Poseidon-Firmament is able to amortize more and more work across workloads.
Poseidon-Firmament Scheduler outperforms the Kubernetes default scheduler by a wide margin when it comes to throughput performance numbers for scenarios where compute resource requirements are somewhat uniform across jobs (Replicasets/Deployments/Jobs). Poseidon-Firmament scheduler end-to-end throughput performance numbers, including bind time, consistently get better as the number of nodes in a cluster increase. For example, for a 2,700 node cluster (shown in the graphs here), Poseidon-Firmament scheduler achieves a 7X or greater end-to-end throughput than the Kubernetes default scheduler, which includes bind time.
Availability of complex rule constraints.
Scheduling in Poseidon-Firmament is very dynamic; it keeps cluster resources in a global optimal state during every scheduling run.
Highly efficient resource utilizations.
Firmament Flow Network Graph – An Overview
Firmament scheduler runs a min-cost flow algorithm over the flow network to find an optimal flow, from which it extracts the implied workload (pod placements). A flow network is a directed graph whose arcs carry flow from source nodes (i.e. pod nodes) to a sink node. A cost and capacity associated with each arc constrain the flow, and specify preferential routes for it.
Figure 1 below shows an example of a flow network for a cluster with two tasks (workloads or pods) and four machines (nodes) – each workload on the left hand side, is a source of one unit of flow. All such flow must be drained into the sink node (S) for a feasible solution to the optimization problem.
Figure 1. Example of a Flow Network
Poseidon Mediation Layer – An Overview
Poseidon is a service that acts as the integration glue for the Firmament scheduler with Kubernetes. It augments the current Kubernetes scheduling capabilities by incorporating new flow network graph based Firmament scheduling capabilities alongside the default Kubernetes Scheduler; multiple schedulers running simultaneously. Figure 2 below describes the high level overall design as far as how Poseidon integration glue works in conjunction with the underlying Firmament flow network graph based scheduler.
As part of the Kubernetes multiple schedulers support, each new pod is typically scheduled by the default scheduler, but Kubernetes can be instructed to use another scheduler by specifying the name of another custom scheduler (in our case, Poseidon-Firmament) at the time of pod deployment. In this case, the default scheduler will ignore that Pod and allow Poseidon scheduler to schedule the Pod to a relevant node.
Note: For details about the design of this project see the design document.
Possible Use Case Scenarios – When To Use It
Poseidon-Firmament scheduler enables extremely high throughput scheduling environment at scale due to its bulk scheduling approach superiority versus K8s pod-at-a-time approach. In our extensive tests, we have observed substantial throughput benefits as long as resource requirements (CPU/Memory) for incoming Pods is uniform across jobs (Replicasets/Deployments/Jobs), mainly due to efficient amortization of work across jobs.
Although, Poseidon-Firmament scheduler is capable of scheduling various types of workloads (service, batch, etc.), following are the few use cases where it excels the most:
For “Big Data/AI” jobs consisting of a large number of tasks, throughput benefits are tremendous.
Substantial throughput benefits also for service or batch job scenarios where workload resource requirements are uniform across jobs (Replicasets/Deplyments/Jobs).
