sks-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Sks-devel] Tuning


From: Jeffrey Johnson
Subject: Re: [Sks-devel] Tuning
Date: Tue, 11 Feb 2014 11:00:06 -0500

On Feb 11, 2014, at 8:38 AM, Christian Reiß <address@hidden> wrote:

> Hey folks,
> 
> 
> pagesize and ptree_pagesize. These options are used for importing/
> generating the db and have no effect on a running server (or?). What
> would be good pointers in setting those?
> 

There are 2 opposing issues choosing pagesize fro Berkeley DB:

        1) the amount of data transferred on each I/O operation
        2) the locking granularity (locks are per page, not per record)

The performance gains were measured a year or so ago and
are in the archives. From memory, there were only modest
performance gains.

The locking granularity tends to be far more import an a consideration (if 
encountered)t.
Berekely DB can deadlock while recursively accessing data. The memory needed
increases for additional locks as the page size decreases, and so other tunables
may need to increase as pagesize decreases.

Use db_tuner(1) (from modern Berkeley DB versions) to help select an optimum
pagesize:
        http://docs.oracle.com/cd/E17076_03/html/api_reference/C/db_tuner.html

Study db_stat(1) post-mortem output (particularly -Cl or -CA) to verify whether 
deadlock’s are occurring
(instead of claiming “corruption”).

Short answer: make pagesize large (64K) and ptree_pagesize small (512), then
use db_tuner to optimize whatever you wish.

hth

73 de Jeff




reply via email to

[Prev in Thread] Current Thread [Next in Thread]