Re: SATA RAID: Promise vs. 3ware

От: Ron
Тема: Re: SATA RAID: Promise vs. 3ware
Дата: ,
Msg-id: E1HTjJv-0002FR-PL@elasmtp-dupuy.atl.sa.earthlink.net
(см: обсуждение, исходный текст)
Ответ на: Re: SATA RAID: Promise vs. 3ware  (Dave Cramer)
Список: pgsql-performance

Скрыть дерево обсуждения

Vacuum full is slow  (Ruben Rubio, )
 Re: Vacuum full is slow  (Heikki Linnakangas, )
  Re: Vacuum full is slow  (Ruben Rubio, )
   Re: Vacuum full is slow  ("Shoaib Mir", )
 SATA RAID: Promise vs. 3ware  (Ireneusz Pluta, )
  Re: SATA RAID: Promise vs. 3ware  ("Merlin Moncure", )
   Re: SATA RAID: Promise vs. 3ware  (Magnus Hagander, )
  Re: SATA RAID: Promise vs. 3ware  ("Joshua D. Drake", )
  Re: SATA RAID: Promise vs. 3ware  (Dave Cramer, )
 Re: SATA RAID: Promise vs. 3ware  (Dave Cramer, )
  Re: SATA RAID: Promise vs. 3ware  (Ron, )
 Re: Vacuum full is slow  (Scott Marlowe, )

At 02:08 PM 3/20/2007, Dave Cramer wrote:

>On 20-Mar-07, at 1:53 PM, Benjamin Arai wrote:
>
>>This is a little biased but I would stay away from areca only
>>because they have fans on the card.  At some point down the line
>>that card is going to die.  When it does there is really no telling
>>what it will do to your data.

Ummm ?what? fan? The Intel IOP341 (AKA 81341) based ARC-12xx cards
are what people are most likely going to want to buy at this point,
and they are fanless:
http://www.areca.us/support/photo_gallery.htm

The "lore" is that
+3ware is best at random IO and Areca is best at streaming IO.  OLTP
=> 3ware.  OLAP => Areca.
- stay away from Adaptec or Promise for any mission critical role.
= LSI is a mixed bag.


>Well, they are also the only one of the bunch that I am aware of
>that  will sell you 1G of cache.

Actually, it's up to 2GB of BB cache...  2GB DDR2 SDRAMs are cheap
and easy to get now.  I've actually been agitating for Areca to
support 4GB of RAM.


>Plus if you use battery backup sooner or later you have to replace
>the batteries. I use areca all the time and I've never had a fan
>die, but I admit it is a point of failure.

I've had the whole card die (massive cooling failure in NOC led to
...), but never any component on the card.  OTOH, I'm conservative
about how much heat per unit area I'm willing to allow to occur in or
near my DB servers.

Cheers,
Ron



В списке pgsql-performance по дате сообщения:

От: Tom Lane
Дата:
Сообщение: Re: Horrible trigger performance after upgrade 8.0.12 -> 8.2.3
От: Dan Harris
Дата:
Сообщение: Determining server load from client