Hi All > That hypothesis could be easily tested by examining the corresponding strace > output for mysql... I don't know if sqlite3's unaligned log writes are > actually slow (didn't see this info in any of Keith's messages), but if they > are that would suggest the hypothesis is false; it would really put nails in > the coffin if mysql uses aligned log writes. > > (BTW Michael, whatever mechanism you use to reply breaks threading in both > Thunderbird and the list archives at sqlite.org; makes it harder to follow > conversations)
Thank you so much for all the input. I tried 2 things, using only a single connection (as suggested by Clemen) and to change the page size to 32K. Single connection did not reduce the average latency, but did help in reducing the spikes. So, I believe Clemen has helped identify one issue (write concurrencies). Using 32K page size also did help reduce the spikes (not sure logic behind this, have yet to do a strace), but average latency is still there. As expected, combining the 2 solutions would not reduce the average latency, but spikes are not as frequent anymore. Eg I still see 160ms average, and sometimes it spikes to 300ms, but I have not seen a 4000ms transaction. I don't think there is anything more I do about this at this stage, I am getting a 7200rpm drive later today to test and see if that makes a difference. I am expecting the average latency to drop by about 25%. Regards Keith _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users