NETS 212 Notes Couple Lectures What am I supposed to say PDF

Preview

Summary

What am I supposed to say What am I supposed to say What am I supposed to say What am I supposed to say What am I supposed to say What am I supposed to say...

Description

09/07 Heterogeneity of the machines become burdensome Computer cluster != Desktop - Many-similar machines, close interconnection - Often special, standardized hardware (racks, blades) - Usually owned and used by a single organization Power and Cooling - Clusters need lots of power - Large clusters need incredible amounts of cooling Beyond small clusters - If the cluster goes out of control - Build a separate building specifically for compute clusters = data center Datacenter terminology - Rack (review slide again) - Emergency power supplies - Loss of data - Data centers consume a lot of energy - The cost factor for energy - The next level: build a network of data centers Potential issues related to data centerS - Correlated faults - Need to put them in different places - Proximity to customers (physics) - Latency issue (low latency) - 100 milliseconds to send the data packets - Costs - Legal restrictions Issues with classical scaling - Difficult to dimension - Load can vary considerably - People’s pattern getting online; the peaks are driven by human activity - Computation for other stuffs (precomputation & indexing) in the downtime (for humans) - Expensive - Need to many $$$ in hardware - Need expertise - Need maintenance

-

Difficult to scale - Scaling up is difficult - Buying machines take weeks - Scaling down is difficult - Idle machines still consume power - Problematic fixed costs such as construction

Recap: Computing at Scale - Modern applications require huge amounts of processing and data - Clusters and data centers can provide the resources we need - Clusters and data centers are not perfect Utility computing and cloud computing -

-

-

Merit - Economies of scale - Statistical multiplexing; a single pool of power shared amongst users - No up-front commitment (no investment in generator; pay as-you-go) - Scalability; add more within seconds NIST’s definition of cloud computing - On-demand self service - Broad network access - Resource pooling - Rapid elasticity (?) - Measured services Recap - Utility computing (elastic) - The Web (information sharing model_ - Web services are run on clouds (but not all) - The Internet - Network of networks; used by the Web; connects clouds to their customers

Everything as a Service (XaaS) - What kind of service? - Does it offer an application or just resources? - If resources, what level of abstra ction - Three main types - Software as a service (Saas) - Provides an entire application (application, middleware, hardware) - You are still paying for it (your data) - Google Drive, Salesforce - Platform as a service (Paas) - Provides middleware / infrastructure (hardware)

-

-

Third party that manages the application (MS Azure, Amazon SageMaker, Windows Azure, Google Cloud Engine) - End-users are still getting the service - Infrastructure as a service (IaaS) - Amazon EC2 instance, Rackspace Cloud, GoGrid - Virtual machine, blade server, hard disk - Responsible for more stuffs - Who can use SaaS/PaaS/IaaS on the cloud - Public cloud: commercial service; open to anyone - Private cloud: shared within a single organization (internal datacenter of a large company) - Community cloud (certain groups): shared by several simlar organizations - Google’s Gov Cloud Recap: motivation for utility / cloud computing - Elasticity, low price, no up-front costs - Different types of clouds - Still, there are some data not suitable - Something that needs microseconds latency (e.g., trading application) - Privacy / compliance

Virtualization - Take physical machines and sell each customer a virtual machine (VM) with the requested resources - From each customer’s perspective, it appears as if they had physical machines all by themselves - Machine has other parts than CPU and memory so these shared devices might slow down one user’s operation - Dynamic threading?? (How is this different from virtualization?) - Resources are virtualized - VMM (Hypervisor) has translation tables that map requests for virtual resources to physical resources - How do VMMs differ from OS kernels? - Two levels of interaction - How expensive is the VMM? Not so much Benefits of Virtualization - Migration - What if the machine needs to be shut down? (e.g. for maintenance, consolidation) - Physical machine for the VM can change over time - Switch off some VMs to move to a single machine for power efficiency - ** User really doesn’t notice this - Time sharing - Multiple VMs can time-share the existing resources

-

-

- More virtual memory and CPU Isolation (Good & Bad) - A user can’t access another user’s dasta - What if the load suddenly increases? Recap: - Gives cloud provider a lot of flexibility - Convenient for users - Provides security and isolation - But: performance may be hard to predict -...