- 1. * Handling Spike Events Reza Rassool, CTO Kwaai Oak
2. * Kwaai Oak212/13/2013 3. SunMonMINSatKwaai OakTueWedThuFri*
312/13/2013 4. Recovered revenue (x$100) Lost revenue (x$100)
Servers ROI% (x100) Page load time (s) Bounce Rate% (x10) Traffic
(100s)8.007.006.005.004.003.002.001.000.00 0Kwaai Oak1234*5678910
11 12 13 14 15 16 17 18 19 20 21 22 23Hour of the day412/13/2013 5.
* Kwaai Oak512/13/2013 6. Standard cloud elasticity can cope with
slowly* varying loads But what about unexpected spikes?* Kwaai
Oak612/13/2013 7. 300.00* Linux bootlatency results in lost
revenue250.00 Total Traffic (x100) Servers200.00* Cloud
billingBounce rate %unfairly pads charges long after the spike is
over.Cloud Cost150.00100.00Not to
scale50.00500550600650700750800Linux Boot time 15mins Kwaai
Oak712/13/2013 8. * Faster booting nodes see GCE * Sub hour billing
* Google GCE (10 min) * MS Azure (1 min)* Kwaai Oak812/13/2013 9. *
Kwaai Oak912/13/2013 10. * Faster boots * Sub-hour billing * Also *
Timely feedback metrics* User centric performance metrics * Load
test (SOASTA) * Dont be afraid to compute ROI * Lost revenue *
Recovered revenue* Kwaai Oak1012/13/2013
