I’m going to add this site to my blogroll. The author is one of VMWare’s R&D honchos.
Interesting tid-bit on how to DOS Google if you’re so inclined.
Some of the search queries can cause huge IO. One example Jeff gave is “circle of life” as one phase enclosed in double quotation marks. It could incur 30GB I/O before. As I just searched on Google, the phrase is now a song name. I bet it’s now in Google’s cache server.
When there are so many servers to manage, something unique happened. Jeff mentioned an interesting phenomenon called “query of death.” If a query can causes a server to crash, then it can crash all other servers because the software stack is the same. To avoid large scale of crashes, they used canary request which is first sent to one machine. If it’s good, then send it to the rest of machines; otherwise reject the request after failing several times. It of course adds a little delay but far better than large scale system crash. Of course, you need to log down the query and found out why it crashed software as a process of continuous improvement. Jeff didn’t mention this, but I bet Google did that.
So, I wonder what the I/O load is on google if I did a search on:
“Ubuntu Private Cloud Prince William Engagement” or “MySQL Replication Senator McConnell Earmark ban”
Pick any two other totally different subject and combine them into a single query.