CS 744: DATACENTER AS A COMPUTER
Transcript of CS 744: DATACENTER AS A COMPUTER
![Page 1: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/1.jpg)
CS 744: DATACENTER AS A COMPUTER
Shivaram Venkataraman Fall 2019
With slides from Mosharaf Chowdhury and Ion Stoica
![Page 2: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/2.jpg)
ANNOUNCEMENTS
- Assignments - Assignment zero is due! - Form groups for Assignment 1 on Piazza
- Class format - Lecture - Review - Discussion
![Page 3: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/3.jpg)
Scalable Storage Systems
Datacenter Architecture
Resource Management
Computational Engines
Machine Learning SQL Streaming Graph
Applications
![Page 4: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/4.jpg)
OUTLINE
- Hardware Trends - Datacenter design - WSC workloads - Discussion
![Page 5: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/5.jpg)
Why is One Machine Not Enough?
![Page 6: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/6.jpg)
What’s in a Machine?
Interconnected compute and storage Newer Hardware
- GPUs, FPGAs - RDMA, NVlink
Memory Bus
Ethe
rnet SATA
PCIe v4
![Page 7: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/7.jpg)
Scale Up: Make More Powerful Machines
Moore’s law – Stated 52 years ago by Intel
founder Gordon Moore – Number of transistors on
microchip double every 2 years
– Today “closer to 2.5 years” Intel CEO Brian Krzanich
![Page 8: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/8.jpg)
Dennard Scaling is the Problem
Suggested that power requirements are proportional to the area for transistors
– Both voltage and current being proportional to length
– Stated in 1974 by Robert H. Dennard (DRAM inventor)
Broken since 2005 “Adapting to Thrive in a New Economy of Memory Abundance,” Bresniker et al
![Page 9: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/9.jpg)
Dennard Scaling is the Problem
Performance per-core is stalled Number of cores is increasing
“Adapting to Thrive in a New Economy of Memory Abundance,” Bresniker et al
![Page 10: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/10.jpg)
Memory TRENDS
![Page 11: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/11.jpg)
MEMORY TAKEAWAY
Growing +15% per year Data access from memory is getting more expensive !
![Page 12: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/12.jpg)
HDD CAPACITY
![Page 13: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/13.jpg)
HDD BANDWIDTH
Disk bandwidth is not growing
![Page 14: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/14.jpg)
SSDs
Performance: – Reads: 25us latency – Write: 200us latency – Erase: 1,5 ms
Steady state, when SSD full – One erase every 64 or 128 reads (depending on page size)
Lifetime: 100,000-1 million writes per page
![Page 15: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/15.jpg)
SSD VS HDD COST
![Page 16: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/16.jpg)
Amazon EC2 (2014)
Machine Memory (GB) Compute Units (ECU)
Local Storage (GB) Cost / hour
t1.micro 0.615 1 0 $0.02
m1.xlarge 15 8 1680 $0.48
cc2.8xlarge 60.5 88 (Xeon 2670) 3360 $2.40
1 ECU = CPU capacity of a 1.0-1.2 GHz 2007 Opteron or 2007 Xeon processor
![Page 17: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/17.jpg)
Amazon EC2 (2018)
Machine Memory (GB) Compute Units (ECU)
Local Storage (GB) Cost / hour
t2.nano 0.5 1 0 $0.0058
r5d.24xlarge 244 768 104 96 4x900 NVMe $6.912
x1.32xlarge 2 TB 4 * Xeon E7 3.4 TB (SSD) $13.338
p3.16xlarge 488 GB 8 Nvidia Tesla V100 GPUs 0 $24.48
![Page 18: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/18.jpg)
Amazon EC2 (2019)
Machine Memory (GB) Compute Units (ECU)
Local Storage (GB) Cost / hour
t2.nano 0.5 1 0 $0.0058
r5d.24xlarge 768 96 4x900 NVMe $6.912
x1e.32xlarge 2 TB 4 TB 4 * Xeon E7 3.4 TB (SSD) $26.68
p3dn.24xlarge 488 768 GB 8 Nvidia Tesla V100 GPUs 0 $31.21
![Page 19: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/19.jpg)
Ethernet Bandwidth
1998
1995
2002
2017
Growing 33-40% per year !
![Page 20: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/20.jpg)
![Page 21: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/21.jpg)
TRENDS SUMMARY
CPU speed per core is flat Memory bandwidth growing slower than capacity SSD, NVMe replacing HDDs Ethernet bandwidth growing Scale up vs Scale out? (Discussion)
![Page 22: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/22.jpg)
DATACENTER ARCHITECHTURE
Memory Bus
Ethe
rnet
SATA
PCIe
ServerServer
![Page 23: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/23.jpg)
Datacenter Networks
Traditional hierarchical topology – Expensive – Difficult to scale – High oversubscription – Smaller path diversity – …
Core
Agg.
Edge
![Page 24: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/24.jpg)
STORAGE HIERARCHY (v2)
![Page 25: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/25.jpg)
Scale Out: Warehouse-Scale Computers
Single organization Homogeneity (to some extent) Cost efficiency at scale
– Multiplexing across applications and services
– Rent it out!
Many concerns – Infrastructure – Networking – Storage – Software – Power/Energy – Failure/Recovery – …
![Page 26: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/26.jpg)
MAIN COMPONENTS OF WSC
![Page 27: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/27.jpg)
SOFTWARE IMPLICATIONS
Workload Diversity
Reliability
Single organization
Storage Hierarchy
![Page 28: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/28.jpg)
Three Categories of Software
1. Platform-level – Software firmware that are present in every machine
2. Cluster-level – Distributed systems to enable everything
3. Application-level – User-facing applications built on top
![Page 29: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/29.jpg)
BigData
WORKLOAD: Partition-Aggregate
Top-level Aggregator
Mid-level Aggregators
Workers
![Page 30: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/30.jpg)
WORKLOAD: Map-Reduce
Reduce StageMap Stage
![Page 31: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/31.jpg)
VIDEO ENCODING
![Page 32: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/32.jpg)
MACHINE LEARNING
![Page 33: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/33.jpg)
WORKLOAD PATTERNS
![Page 34: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/34.jpg)
DATACENTER VS DESKTOP
Parallelism Available Workload churn Platform homogeneity Fault-free operation
![Page 35: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/35.jpg)
DISCUSSION
Form groups of 4 students Pick up a discussion form per group Fill out responses at https://forms.gle/hhuKktMb5pKkotwc9
![Page 36: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/36.jpg)
Discussion
Scale-up vs Scale-out
![Page 37: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/37.jpg)
DISCUSSION
Differences between web-search and MapReduce
![Page 38: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/38.jpg)
DISCUSSION
Microsoft Word vs. online document editor like Google Docs
![Page 39: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/39.jpg)
DISCUSSION
![Page 40: CS 744: DATACENTER AS A COMPUTER](https://reader035.fdocuments.in/reader035/viewer/2022062306/62a68c31b8eb4c30ca250e63/html5/thumbnails/40.jpg)
NEXT STEPS
9/12 class on Storage Systems Assignment 1 out Thursday