AWS - an introduction to bursting (GP2 - T2)
Download as PPTX, PDF15 likes12,133 views
The document introduces AWS's burstable EC2 instance types (t2) and EBS volume type (gp2), both designed for cost-effective performance with the ability to exceed baseline performance during high-demand periods using 'credits.' It outlines the characteristics, use cases, and cost comparisons for these services, highlighting their suitability for low-traffic applications and development environments. Additionally, it provides guidance on pricing, performance metrics, and best practices for utilizing these new features effectively.
![Is T2 any faster than T1?
Doorkeeper migrate their RDS database
Faster CPU means lower load More memory means better caching
“With the t2.micro […] we now spend ~35ms on SQL queries - with higher overall rpm,
compared to ~100ms before.”
Michael Reinsch, Co-Founder Doorkeeper Inc](https://image.slidesharecdn.com/aws-anintroductiontoburstinggp2-t2-140821071414-phpapp02/85/AWS-an-introduction-to-bursting-GP2-T2-14-320.jpg)
AWS - an introduction to bursting (GP2 - T2)
- 1. AWS – an introduction to bursting New EC2 (T2) and EBS (GP2) volume types 2014-08-21 Amazon Data Services Japan KK Rasmus Ekman
- 2. So, what is a “burst”? Burst NormalNormal • When you have a low average baseline, with a large temporary spike, or burst, of performance • Example: even though the speedometer of my motorcycle maxes out at 250KMH, I rarely drive at that speed, but it is certainly nice to have the option to do so when the time and the circumstances are right. Most of the time I am using just a fraction of the power that is available to me.
- 3. New low cost AWS burst services • New EC2 instance family, T2, and EBS volume type, GP2, engineered with bursting in mind • Common use cases – Low traffic web servers – Dev. environments(build servers etc) – Remote desktops – Small databases • Extremely high cost- performance Computing T2 New EBS volume type GP2 Storage New instance family
- 4. It’s all about the credits • Credits are “saved” when resources are under low load, and used during bursts CPU creditsCPU usage = Spending Savings
- 5. Introduction of service specifics Computing T2New instance family
- 6. What is “T2”? • T2 instance types are a special breed that differ from normal families, such as M3, R3 and C3, by being able to burst CPU using CPU credits at a very low cost • Intel Xeon 2.5GHz CPU（with turbo up to 3.3GHz） • T2 instances provide a baseline level, CPU utilization, of CPU performance. The baseline is a percentage of the full core performance • The very lowest cost instance type, t2.micro, is part of the free tier! • On demand rate starts at $0.013/hour, about $9.50/Month model vCPU memory storage Baseline performance t2.micro 1 1GB EBS only 10% t2.small 1 2GB EBS only 20% t2.medium 2 4GB EBS only 40%
- 7. In other words, T2 is a small instance type, with very high cost-performance, that can burst
- 8. T2 is all about understanding CPU credits Instance type Initial (boot) credits Credits earned per hour Base performance (CPU utilization) Maximum CPU credit balance t2.micro 30 6 10% 144 t2.small 30 12 20% 288 t2.medium 60 24 40%(*) 576 To simplify, １ CPU credit ＝ 1 minute of burst burst＝usage above the baseline performance Details/(*)：http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/t2-instances.html
- 9. Monitor CPU credits through CloudWatch
- 10. Common use cases for T2 • Development environment • Code repository • Low traffic web application • Small scale database, etc etc If you find yourself constantly running out of credits, move to a higher instance type (e.g. M3) Hint→
- 11. Price（Tokyo region） • t2.micro – Monthly cost (Linux) – $14.6 (On Demand) – $10.7 (1-year Heavy RI) – $7.3 (3-year Heavy RI) • t2.micro – Monthly cost (Windows) – $18.3 (On Demand) – $12.8 (1-year Heavy RI) – $9.4 (3-year Heavy RI) • t2.small/medium will run 2x and 4x With a 3 year RI, you can get a Windows server under $10/Month!
- 12. On-demand VS reserved instance (RI) 0 100 200 300 400 500 600 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 t2.micro t2.micro OD t2.micro 1YHURI t2.micro 3YHURI USD 月 On-demand 1 year RI 3 year RI On-demand 1 year RI 3 year RI
- 13. Recommended migration path t1.micro t2.micro m1.small t2.small m1.medium t2.medium
- 14. Is T2 any faster than T1? Doorkeeper migrate their RDS database Faster CPU means lower load More memory means better caching “With the t2.micro […] we now spend ~35ms on SQL queries - with higher overall rpm, compared to ~100ms before.” Michael Reinsch, Co-Founder Doorkeeper Inc
- 15. T2 pointers • T2 only supports HVM – Same as R3/I2 • T1 only supports PV, meaning you can’t just do an instance modification (1 click migration) to T2 • T2 only supports VPC, so if you are still on classic, time to consider a move to VPC • A general shift towards HVM (Intel CPU virtualization feature) • When you reboot, your CPU credits are reset to 0
- 16. Introduction of service specifics New EBS volume type GP2 Storage
- 17. Existing EBS volume types Magnetic (standard) Provisioned IOPS (PIOPS) Type Hard disk base SSD base Capacity fee Fee per GB Fee per GB IOPS capacity fee - Fee per provisioned IOPS IO request fee Fee per million IO requests - Performance 100 IOPS on average with the ability to burst to hundreds of IOPS Consistently performs at provisioned level, 4000 IOPS maximum
- 18. Introducing EBS General Purpose (SSD) – GP2 a new contender arrives Magnetic (standard) General Purpose SSD Provisioned IOPS (PIOPS) Type Hard disk base SSD base SSD base Capacity fee Fee per GB Fee per GB Fee per GB IOPS capacity fee - - Fee per provisioned IOPS IO request fee Fee per million IO requests - - Performance 100 IOPS on average with the ability to burst to hundreds of IOPS The ability to burst to 3,000 IOPS maximum, with a base performance of 3 IOPS/GiB Consistently performs at provisioned level, 4000 IOPS maximum
- 19. In other words, GP2 is… A new SSD based EBS volume type that can burst, and has a very simple fee structure with a high cost- performance
- 20. GP2 details • A new SSD based EBS volume type • Base performance of 3 IOPS per GB 100GB volume gives 300 IOPS, and 500GB gives 1500 IOPS baseline etc. • Can burst up to 3000 IOPS Very good fit for boot volumes, and volumes that have short periods of heavy usage The burst time span is governed by I/O credits (details later) • Simple fee structure $0.12/GB per Month (Tokyo region) No IO request or provisioning fees! details：http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/EBSVolumeTypes.html
- 21. Base performance and volume size Performance of a General Purpose(SSD) volume (continuous IO load with various volume sizes)• When IO capacity over the baseline is requested, bursting will happen • When credits have been exhausted, the baseline of 3 IOPS per GB is maintained • With a 1TB volume you will get a 3000 IOPS baseline IOPS Minutes 500GB 1TB 100GB
- 22. A cost-performance overview Low IO usage High IO usage • Magnetic is marginally cheaper than GP2, but GP2 can burst up to 3000 IOPS without additional cost • During high load scenarios GP2 is much more cost effective than PIOPS • Magnetic cannot provide the performance needed in this scenario • PIOPS will guarantee a stable IOPS 100 IOPS Lowest cost
- 23. General purpose SSD (GP2) tips • 5 to 50 percent reduction in IOPS when you first access each block of data on a newly created or restored EBS volume, so pre-warm the volume before using For best performance, access all blocks before using For new volumes, write to all blocks ※With Linux run the dd command (of=/path/to/devfile), and in Windows do a full format When restoring a snapshot, read all blocks ※With Linux run the dd command (if=/path/to/devfile), and in Windows use “dd for windows” • Use EBS-Optimized Instances By using a separate network for EBS, you will get more stable IO throughput • To convert your old EBS volumes to GP2, take a snapshot From an EBS snapshot you can restore to a different volume type, such as GP2 Pre-warming details: http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/ebs-prewarm.html
- 24. EBS comparison summary Magnetic (standard) General purpose SSD Provisioned IOPS (PIOPS) Use case • Very low usage scenarios. • Consider S3 for backups • OS volume • Virtual desktops • Development and test environments • Very high loads that require extremely stable IOPS performance • Large scale databases perform ance 100 IOPS on average with the ability to burst to hundreds of IOPS The ability to burst to 3,000 IOPS maximum, with a base performance of 3 IOPS/GiB Consistently performs at provisioned level, 4000 IOPS maximum
- 25. Summary
- 26. Summary • Two new low priced burstable services – Computing – T2 family – Storage – general purpose SSD (GP2) • CPU and IO bursts are governed by credits • Very high cost-performance • Good fit for use cases that can allow for variable performance
- 27. appendix
- 28. IOPS and I/O Credits detailed example IOPS I/O Credit Initial I/O Credit is 5,400,000 As long as IO credits remain, performance will be 3000 IOPS While bursting IO credits will gradually be consumed One credits are depleted bursting will stop Once credits reach 0, the baseline performance will be excercised Once load is under baseline, credits will increase When increasing the load again after credits have been restored, burst will happen 分 Baseline performance (gp2 500GB example)