Re: Cache battery and Raid Controller Design WAS: Re: Rép. : Question of a Newbie to POWER Systems

If you are watching the QSYSOPR and QSYSMSG (if it exists) message queues at all, you will get several warnings at least 90 days prior to the cache being shut down. If that happens, it's because your not watching the system.

Jim Oberholtzer
CEO/Chief Technical Architect
Agile Technology Architects, LLC


On 3/30/2011 8:54 AM, Bryce Martin wrote:

Yeah, I wouldn't advocate it, but at least you know its there if the
system decides to end caching in the middle of the day when the system is
needed the most. What a great day of learning for me. haha.


Thanks
Bryce Martin
Programmer/Analyst I
570-546-4777



Vern Hamberg<vhamberg@xxxxxxxxxxx>
Sent by:midrange-l-bounces@xxxxxxxxxxxx
03/30/2011 09:45 AM
Please respond to
Midrange Systems Technical Discussion<midrange-l@xxxxxxxxxxxx>


To
Midrange Systems Technical Discussion<midrange-l@xxxxxxxxxxxx>
cc

Subject
Re: Cache battery and Raid Controller Design WAS: Re: Rép. : Question
of a Newbie to POWER Systems






Bryce

I have sort of done what you speak of here, but for a new battery pack.
It somehow didn't register - or I didn't do the procedure correctly.
Anyhow, forcing the error condition, then doing the procedure again with
the same battery, did reset things. This was when I felt that I knew it
was a good battery pack. This was also a box that is not in constant
use, and we are a software house, so our needs are different from
regular customers.

The aging of the battery is based on time and on environment -
specifically, temperature. If it is warmer, IBM thinks the life will be
shorter. Some of the packs actually have a connection for a temp sensor,
as I recall. You can get a pack through non-IBM sources, but they will
not have the temperature bit.

But if the pack has been there a while, I do not recommend faking out
the system. I did it when I put in a spanking brand-new one. I got lucky
- it's still working fine, no slowdown, etc.

Vern

On 3/30/2011 8:00 AM, Bryce Martin wrote:

> What you and Jim say makes sense. Its more of IBM making you go from
> cache to synchronous when they feel the cache battery might be

unreliable

> instead of waiting for it to fail and you possibly losing data.
>
> I was going to suggest a way to somehow override the system so that you
> could get your performance back while waiting on a replacement, but then

I

> remembered hearing that you can reset the clock to make the system think
> it was replace correct? Because doesn't it just work on some sort of
> clock hours??
>
>
> Thanks
> Bryce Martin
> Programmer/Analyst I
> 570-546-4777
>
>
>
> Vern Hamberg<vhamberg@xxxxxxxxxxx>
> Sent by:midrange-l-bounces@xxxxxxxxxxxx
> 03/30/2011 08:49 AM
> Please respond to
> Midrange Systems Technical Discussion<midrange-l@xxxxxxxxxxxx>
>
>
> To
> Midrange Systems Technical Discussion<midrange-l@xxxxxxxxxxxx>
> cc
>
> Subject
> Re: Cache battery and Raid Controller Design WAS: Re: Rép. : Question
> of a Newbie to POWER Systems
>
>
>
>
>
>
> Bryce
>
> Interesting question - seems to me that the cache battery is a backup -
> it preserves anything in the cache in the event of a power outage. The
> system could and probably does use the regular power of the system
> during normal operations.
>
> The matter of slowing down when the cache battery's life is running out
> - I think this happens for a different reason - if the battery can't
> keep data in the cache, then you can't rely on the cache. So writes have
> to be done synchronously, not through the cache, because the backup is
> not trustworthy. This slows it down.
>
> I've not thought of this at all, so my ideas can be far off base!
>
> Vern
>
> On 3/30/2011 7:32 AM, Bryce Martin wrote:

>> Jim Said....
>> "Write cache has a massive impact on I/O performance. The more write
>> cache the better. Ask anyone who has lost the batteries on the raid
>> card what happens when write cache goes away. Ugly things happen."
>>
>> This got me thinking about RAID Card design....
>>
>> Why the heck isn't the cache batter a backup, not a primary power

> source?

>> If the cache batter goes shouldn't the controller still have enough
>> electrical input to keep its cache alive? This just seems like poor
>> design, and a major oversight. Maybe I don't understand hardware

design

>> (that is probably the case), but I'm failing to see what the purpose is

> of

>> the battery vs straight electrical input from the system? I would

think

>> you'd want a cache battery in the case of a hardware failure so you

> don't

>> lose data, but I would think that it should be a backup source....
>>
>> Thanks
>> Bryce Martin
>> Programmer/Analyst I
>> 570-546-4777