Back of the envelop estimations
Hard Drives
- HDD - 100 MBps
- SSD - 500 MBps
- Memory Card - 1GBps
Networks
- 4G/LTE - 30MBps to 100MBPS
- LAN - 1 GBPS
- 5G - to be introduced
- 40Gbps (5GBps - GigaBytes per second) - network throughput for a good data
center
- if i have a file of 5GB, it will take a second to transfer the file from one one node to the other
Assume a good quality image of 1 MB
- Thumbnail is 1/10th of the actual image -> 100 KB
| HORIZONTAL | VERTICAL | Notes |
|---|---|---|
| L1 cache reference (per core) | 0.5 ns | 6 Lakh times faster than RAM |
| L2 cache reference | 7 ns | 14~15 times slower than L1 |
| Read 1MB sequentially from disk | 30ms | |
| Write 1MB into the disk | 2 times than Read approximately (for HDD), for SSD, 1.2 times | |
| Read 1MB sequentially from Memory | 300μs (250μs) | 1 lakh times faster than disk |
| Send 1MB over 1Gbps network | 10ms | \[\frac{1\times 8 * 10^6}{10^9} \approx 10\times 10^{-3}\] |
| Read 1MB over 1Gbps network | 10ms | |
| Compress 1MB using Zippy | 10ms | for 1KB, \(\frac{10 * 10^{-3}}{10^{-3}} = 10μs\) |
| Compression ratio | ||
| Round trip within same Data center | 150μs | |
| Round trip CA -> Netherland -> CA | 500ms |
- Compress data before sending over internet
QPS - Queries per second
Peak QPS = 2 to 5 times QPS
Time Estimations
1 Day has $ \textrm{86400 seconds} = 24\times 60\times 60 \approx 90K secs $
100 Million load per day
$ \text{Load per second} = \frac{100 \times 10^6}{90 \times 10^3} = \frac{100{,}000{,}000}{90{,}000} = 1111.11\overline{1} \approx 1200 $
Denominator is slightly increased to 90K so for lesser denominator the output will increase
Estimates
- DAU : Daily Average Users eg. 300 million MAU
- Usage per DAU = 10 -25% make a post (QPS : Queries per second)
- Scale Factor : estimate how much higher the traffic would peak, compared to the average
Latency Numbers :
Billion is $ 10^9 $ nanosecond is 1 billionth of a second
Million is $ 10^6 $ microsecond is 1 millionth of a second
Levels of Availability
| Availability (%) | Downtime per Year | Examples of Suitable Applications |
|---|---|---|
| 99.00% (Two 9’s) | ~3.65 days | Non-critical applications, occasional downtime acceptable |
| 99.90% (Three 9’s) | ~8.76 hours | Internal business applications, moderate uptime requirement |
| 99.99% (Four 9’s) | ~52.6 minutes | Customer-facing applications, e-commerce sites |
| 99.999% (Five 9’s) | ~5.26 minutes | Mission-critical applications, financial services |
| 99.9999% (Six 9’s) | ~31.5 seconds | Highly critical systems, emergency services |