[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: syncrepl questions

--On Thursday, September 25, 2003 1:00 PM -0700 Quanah Gibson-Mount <quanah@stanford.edu> wrote:

Well, how's refreshAndPersist's performance at the intial refresh stage ?

Hard to judge at this point -- It starts getting the information from the master for 30 seconds, then the connection is closed, after which point it never opens a new connection. I'll remove the idletimeout on the master and test some more.

Okay, in the initial stage, here is what happens:

It hits the master, which jumps to about 20% of the CPU.
slapd starts with a VSIZE of 2058MB (for my 2GB BDB cache), and a RES size of 3872KB. The master is using HDB as its database. As the query progresses, the sizes change to 2143MB VSIZE and 953MB RES size. So the resident size on the master grows by 950MB. After the replica stops querying, the slapd process drops to .04% of the CPU or so.

On the replica, slapd starts at:
2059MB VSIZE and 5240KB RES. After the initial query to the master completes, slapd grows to about 35% of the CPU. RES memory size grows to about 1GB during this time. After this is done processing, slapd idles, dropping to about 0.00% of the CPU, with a RES size of 5360KB. The VSIZE goes stays the same. The replica has BDB as its database.


Quanah Gibson-Mount
Principal Software Developer
ITSS/TSS/Computing Systems
Stanford University
GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html