Talk:Seqlock

From Wikipedia, the free encyclopedia

The referenced PDF file, http://www.sgi.com/products/software/linux/downloads/sync_linux.pdf, is now a dead link. It is hard to tell whether the presentation titled "Effective Synchronization on Linux/NUMA Systems" at http://www.gelato.org/pdf/may2005/gelato_may2005_numa_lameter_sgi.pdf is the PDF file in question, but I would guess not. It is the closest match I was able to find using Google, though.

[edit] Clarification on how the reader ensures integrity of read data?

Question: per article, would not just checking evenness after the read be not enough, for the reader? For example, in a situtation where the reader reads the counter (say its 2), performs the read while the writer modifies the data, checks the counter again - it's 4. The data read is not consistent, but the counter is still even :). I hope I'm not insane or anything... So, is that something that needs to be clarified in the article or am I just plain wrong?

---

Andrei Warkentin (andrey.warkentin@gmail.com)

I have fixed the article to match LWN's description (Corbet is an acknowledged authority on the kernel APIs), you are quite right that as described it wouldn't work. 82.69.171.90 (talk) 12:12, 7 February 2008 (UTC) That is, as /previously/ described, of course. I think the description I wrote is accurate. 82.69.171.90 (talk) 12:13, 7 February 2008 (UTC)