I agree, HTM can be used in a number of places, as indicated by the section title "Where is HTM Most Likely to be Helpful?" The contents of that section are thus the areas where I believe HTM has the most promise, not the only places that HTM could be used.
On your point about RCU and hazard pointers vs. TM, it is important to note that this is not necessarily a strict either-or decision. For an example using both RCU and STM on a red-black tree, please refer to http://www.usenix.org/event/hotpar11/te ch/final_files/Howard.pdf. This paper gives a clear example of how using multiple tools in concert can give superior results.
But what is your definition of "transparent"? Any reasonable definition I have been able to come up with has RCU being either more or less transparent, depending on the workload.
On your point about RCU and hazard pointers vs. TM, it is important to note that this is not necessarily a strict either-or decision. For an example using both RCU and STM on a red-black tree, please refer to http://www.usenix.org/event/hotpar11/te
But what is your definition of "transparent"? Any reasonable definition I have been able to come up with has RCU being either more or less transparent, depending on the workload.
Comment Form