Re: Enterprise Manager warns of blocking on LGWR

From: Mladen Gogala <gogala.mladen_at_gmail.com>
Date: Fri, 8 Dec 2017 19:46:48 -0500
Message-Id: <20171208194648.aef234fd033d46715673774b_at_gmail.com>


On Fri, 08 Dec 2017 09:16:30 +0000
Norman Dunbar <oracle_at_dunbar-it.co.uk> wrote:

> Be safe, duplex anyway, after all 'disc is cheap' as we are always told, and cpu isn't really a problem, especially if the disc controllers do it for you.

On low end Linux machines and with low-end Q-Logic cards, disk controllers do not really do it for you. It depends on the hardware. And PC architecture doesn't really have smart channels like the AIX boxes, so it is possible to saturate your I/O rather quickly. Doubling your redo log I/O may be a really bad idea with an active OLTP database. With PC architecture, you usually have CPU, memory, network and disk controllers, all on the same bus. Unless your disk controllers are smart enough to do the entire FC handshake for you and put the result directly in memory, without generating too many interrupts along the way, you may be surprised in a rather unpleasant way. The same applies to iSCSI.

-- 
Mladen Gogala
Oracle DBA
Tel: (347) 321-1217
--
http://www.freelists.org/webpage/oracle-l
Received on Sat Dec 09 2017 - 01:46:48 CET

Original text of this message