Test 1 Flashcards
What are the Pros and Cons of Full Virtualization?
Full Virtualization
Pros:
No source code modification required
Easy to implement
Feasible for all CPU architectures
Cons:
Latency from OS/HW emulation
Hard to provide real-time guarantees
What are the pros and cons of paravirtualization?
Paravirtualization
Pros:
Better performance compared to full virtualization
25-75% less RAM usage
2-20% less disk I/O
10% less network I/O
7% less CPU usage for privileged instructions
Cons:
Requires OS modification
Poor portability
Not compatible with off-the-shelf or legacy OS versions
What is a rack?
A large amount of servers mounted and connected together
What is a cluster?
A collection of server racks
How are things stored in data centers?
Through a combination of SSDs and HDDs
What is the main use case of SSD?
Accessing data
What is the main use case of HDD?
Storing data for long term use
What is Directed Attached Storage?
Storage directly connected to the server
What are Network Storages?
Storage connected to a cluster level switch
What are some attributes of Directed Attached Storage?
Backup (at least three) is managed by a distributed file system, low cost but low reliability
What are some attributes of Network Storage?
Contains redundancy and replication mechanisms
What is the goal of data center software?
To maximize the locality of communication and data relative to the rack
Where are data centers located?
Close to internet backbone optical fibers
Areas with cheap and reliable electricity
Areas with low property tax rates
Areas with high stability in the country
What is CAPEX?
Capitial Expenditure which includes the building, power, cooling infrastructure, and initial IT equipment
What is OPEX?
Operational Expenditure which includes, buying replacement equipment, electricity, and salaries
What takes up the most of a Data Centers power?
CPUs, DRAM, Disks, and Networking
Why does power consumption in Data Centers matter?
It costs alot of money
What percentage of global energy is used by data centers?
1%
How to quantify Energy Efficiency?
PUE Power Usage Effectiveness
1+Non IT Equipment Power/ IT Equipment Power
Which is better a higher PUE or lower PUE
Lower PUE 1 is ideal
What is held in the HW layer of Google Data Centers?
Servers, Storage, Networks, Accelerators, Physical Infrastructures
What is held in the Virtualization layer of Google Data Centers?
Bios, MGMT Controller, Drivers, Kernel
What is held in the Resource MGMT layer of Google Data Centers?
BORG Master, BORGLET, Chunkserver
What is held in the Cluster Infrastructure layer of Google Data Centers?
Mapreduce, Spanner, Colossus, Remote Procedure, Auto Scaling