operationally, disk subsystems trump software raid for primary dataset
storage:
battery backed cache - safety first.
portability - plug it into any other host, or more than one host on the scsi
bus (fc-al loop).
raid on a pci card or raid on the motherboard need not apply less they sport
battery and most don't. worst of both worlds imho, no protection of software
and the higher cost of hardware.
often software raid is used to mirror the boot disks for transparent boot
disk failover, while the primary dataset is on a hardware raid on a scsi bus
with two hosts for transparent host failover.
but the question, if i can restate it, was ``which is faster?''
-high algorithm overhead and slower devices
-low algorithm overhead and faster devices
did i miss some nuance of the original question? it seems obvious that lower
overhead and faster devices will be faster.
but it doesn't matter which one is faster. operationally, as long as the
performance is adequate, real disk subsystems are always preferred for
primary dataset storage.
----- Original Message -----
From: "Sander Steffann" <sander@steffann.nl>
To: <chris@ruprecht.org>; "Nikolaus Dilger" <nikolaus@dilger.cc>
Cc: <mallah@trade-india.com>; <pgsql-admin@postgresql.org>
Sent: Monday, November 25, 2002 2:43 AM
Subject: Re: [ADMIN] H/W RAID 5 on slower disks versus no raid on faster
HDDs
: Hi,
:
: > In a production environment I would always favor some
: > kind of error protection. Either RAID 5 or RAID 1
: > (mirroring). A hardware RAID controller is faster than
: > software RAID.
:
: Considering the speed of CPU's and busses these days, software RAID can be
a
: lot faster than hardware RAID in many cases. I prefer hardware RAID myself
: because of the simplicity (the OS doesn't need to know about the RAID
: configuration), but I hear good results from people running on software
: RAID.
:
: Hope this helps you.
: Sander.
:
:
: ---------------------------(end of broadcast)---------------------------
: TIP 6: Have you searched our list archives?
:
: http://archives.postgresql.org
: