Professional Documents
Culture Documents
Cloud Computing and Distributed Systems
Cloud Computing and Distributed Systems
- Black-box only uses data from outside the VM, while Gray-box accesses OS stats and
application logs.
- By incrementing the Xen hypervisor to provide domain-0 with access to CPU scheduling
events.
- Dom 0.
22. How can the lower bound on request rate be determined for the virtual EL
server?
- By substituting service time, variance of inter-arrival and service times into an equation.
24. How should the CPU share of a VM be adjusted to service peak workload?
- By multiplying the estimated peak arrival rate and mean requested file size.
26. What are the stages involved in live VM migration as per the notes?
- Pre-Migration, Reservation, Iterative Pre-Copy, Stop-and-Copy, Commitment, Activation.
- To suspend the running OS instance, transfer CPU state and memory pages.
31. What takes place during the Activation stage of live VM migration?
32. How does the hotspot mitigation algorithm determine VMs to migrate?
- Trap-and-Emulate
36. What problem arose with Trap-and-Emulate in x86 Pre 2005 systems?
- Performance improvement
41. How much Guest OS code may need modification in Para virtualization?
- Less than 2%
46. What technologies were used by Yahoo! and Facebook for big data
processing?
- Hadoop and Pig for Yahoo!; Hadoop and Hive for Facebook
- Public clouds provide service to any paying customer; private clouds are accessible only to
company employees.
50. What factors are considered in deciding to outsource or own a cloud service?
- Efficient, isolated duplicate of a real machine, allowing multiple OSs to run concurrently on the
same physical machine.
56. How does Xen track events to determine the duration of each virtual machine
within a measurement interval?
- Xen tracks events to determine the duration of each virtual machine within each measurement
interval.
- Domain-0 in Xen implements the network interface driver for network monitoring.
58. How does Xen enable multiplexing of virtual interfaces onto the physical
network interface?
- Xen enables multiplexing by attaching virtual interfaces to the virtual firewall-router (VFR).
59. What challenges are associated with black-box monitoring of memory in Xen?
- Challenges include Xen allocating memory to VM and the need for OS management within
each VM.
60. How does Xen determine which memory pages are dirtied during migration?
- Xen uses shadow page tables for migration to determine which pages are dirtied.
- Black-box monitoring is useful when it is not feasible to gather usage statistics inside a VM.
62. What environment allows OS-level statistics and application logs gathering?
- Corporate data centers allow gathering of OS-level statistics and application logs.
63. What is gray-box monitoring in the context of virtual server monitoring?
- Gray-box monitoring involves installing a monitoring daemon within each virtual server.
64. What interface does the monitoring daemon in Linux use to gather OS-level
statistics?
- The monitoring daemon in Linux uses the /proc interface to gather OS-level statistics.
65. How does gray-box monitoring enable explicit detection of SLA violations?
- Gray-box monitoring directly monitors application-level statistics for SLA violation detection.
66. What does a hotspot indicate in the context of virtual server resources?
67. What estimates are needed before resolving a hotspot through migrations?
68. How does hotspot detection differ in black-box and gray-box approaches?
- Black-box approach flags hotspots based on aggregate CPU or network, while gray-box
detects explicitly.
69. What are the three key models for device virtualization?
70. In the Passthrough Model, how does the VMM handle device access
permissions?
- Latency of device operations and device driver ecosystem complexities in the hypervisor.
72. How does the Split-Device Driver Model split device access control?
73. What does the lecture define and discuss regarding virtualization
approaches?
- Monitoring and detecting hotspots by determining new mappings and initiating migrations.
75. What are the two approaches discussed for hotspot mitigation in VM
migration?
78. What are the causes of dynamic workload fluctuations in data center
workloads?
80. How has the nature of data changed from MBs to PBs in cloud computing?
- From MBs to TBs, PBs, and XBs.
81. Name some of the cloud programming paradigms mentioned in the notes.
- MapReduce/Hadoop, NoSQL/Cassandra/MongoDB.
85. Which cloud service classification offers access to software services when
needed?
- Get access to computing and storage infrastructure coupled with a software platform.
87. What measuring metric indicates good power efficiency in a data center?
- Cool air moved through the system by using purified water and air intake.
92. What is the extreme approach for hotspot detection with n=k=1?
- Lower thresholds imply aggressive migrations; higher thresholds imply higher utilizations.
- It predicts future values using prior observations and time series statistics.
- The hotspot detector updates predictions and checks for threshold exceedance.
- When the resource needs exceed the estimated peak CPU, network, and memory
requirements.
99. What is used to estimate peak CPU and network bandwidth needs?
100. Who gave the slogan 'The network is the computer' in 1984?
- The cloud
- Compute nodes, switches, storage nodes, and front-end for job submissions
- Processor speed
108. What law indicates that network bandwidth has doubled each year?
- Gilder's law
109. What was the bandwidth mostly in 1985 according to the notes?
110. What was the disk capacity comparison between today's PCs and a 1990
supercomputer?
112. How did MIT's Fernando Corbató envision computer facility operation in
1965?
113. What business model does utility computing focus on according to the
notes?
114. What major computing paradigms are composable with QoS and SLAs?
115. What method is used to estimate future peak needs based on recent past
usage?
- Computing a high percentile (e.g., the 95th percentile) of the CPU and network bandwidth
distribution.
116. How are peak resource needs estimated in the black-box provisioning
approach?
- Peak needs are estimated by observing the tail of the CPU and network bandwidth usage
distribution.
117. In the black-box approach, how is peak memory need estimation different
from CPU and network needs?
- Memory provisioning is based on observed swap activity rather than high percentiles of usage
distribution.
119. How is the peak arrival rate estimated in the gray-box provisioning
approach?
- By summing the serviced and dropped request rates and identifying a high percentile in the
distribution.
120. Which type of applications can be modeled as G/G/1 queuing systems for
estimating peak CPU needs?
121. What queuing theory result is used to estimate peak CPU needs in G/G/1
queuing systems?
- Utilizing a queuing theory result with mean response time, mean service time, and request
arrival rate.
- The performance impact is significant due to Xen intercepting all memory accesses for the
migrating domain.
123. How can the total migration time and performance impact on applications
be minimized during VM migration?
- Use Volume to find the most loaded servers and target highly loaded servers first based on
multiple resource dimensions.
125. What is an approach discussed in the lecture to rapidly detect and eliminate
hotspots during VM migration?
126. How can virtual machine migration be leveraged in dynamic data center
provisioning based on the conclusion from the course notes?
- As a viable tool for dynamic data center provisioning by rapidly detecting and eliminating
hotspots.
127. According to the course notes, how can gray-box information improve
performance in some scenarios?
- Brief introduction to Cloud Computing and aspects of Why Clouds, What is a Cloud.
129. How have evolutionary changes impacted distributed and cloud computing?
131. According to Gartner in 2009, what was the forecast for cloud computing
revenue by 2013?
- Expected to exceed $150 billion.
- AWS, Microsoft Azure, Google Compute Engine/AppEngine, Rightscale, Salesforce, and more.
134. How can companies save time and money using AWS according to the text?