On 4/5/21 10:23 PM, Robert Brockway wrote:
On Thu, 1 Apr 2021, Jean St-Laurent via NANOG wrote:
What happened is that it would create a kind of internal DDoS and they would all timed out and give a weird error message. Something very useful like Error Code 0x8098808 Please call our support line at this phone number.
If only there was a way to address the Thundering Herd problem before the cloud. :)
This simple change to add 3 lines of code to add a random artificial boot penalty of few seconds, completely solve the problem.
Bingo. Now, the trick is to catch this before it causes an self-DDoS.
This is a problem that has been recognised for decades and this is unfortunately a good example of how operational experience is still not being distributed properly. Too many managers think that operational work is obvious and just a result of common sense. It isn't.
Same problem as disk drives powering up simultaneously in datacenters. SCSI drives have (had?) a random delay mechanism to distribute the initial power surge over a few seconds.