-
1. Re: JournalCopier
clebert.suconic Oct 28, 2009 9:55 AM (in response to timfox)The read method is the same we use on load. The read isn't going to be "very slow" I think.
I also thought about doing it this way because the backup won't need to load the journal at the end of the copy. The Journal will be ready to be used as soon as the last add is sent.
I'm also only sending the valid records. I'm not sending any deleted records... the same process as the compacting.
I could run a test to validate the performance if you want. -
2. Re: JournalCopier
timfox Oct 28, 2009 9:57 AM (in response to timfox)Are you sending each record across the wire in it's own packet?
-
3. Re: JournalCopier
clebert.suconic Oct 28, 2009 10:00 AM (in response to timfox)Yes.. Allthough I could optimize it.
-
4. Re: JournalCopier
timfox Oct 28, 2009 10:03 AM (in response to timfox)Standard ethernet MTU is 1500 bytes.
So if average record size is < 1500 bytes, then you're mostly sending empty bytes across the network.
This will hugely affect performance. Make sure you set tcp no delay to false to get around this.