ADSM-L

Re: MIRRORING DILEMMA

1999-11-01 22:48:00
Subject: Re: MIRRORING DILEMMA
From: William Dias <wdias AT US.BNSMC DOT COM>
To: ADSM-L AT VM.MARIST DOT EDU <ADSM-L AT VM.MARIST DOT EDU>
Date: Tuesday, November 02, 1999 3:48 PM
>Mirror mirror of my data,
>Which one will let me recover later!
>
>I have gone back and read the My mirroring is better than your Mirroring
and I'm
>sorry its all nonsense!   First off, there is no such thing as hardware
mirroring.
>If you are using AIX mirroring (exact copies on multiple drives)  the
mirroring is
>done by the device driver code and LVM.   If you are using RAID5 the
mirroring is
>done in the RAID adapter using both hardware and micro code. (Micro code is
an old
>IBM term for software used on an adapter card).  The closer you are to the
hardware
>the more efficient you can make the operation.  The better you understand
the
>hardware the greater the chance of recovering the data.  ADSM is an
application far
>removed from the base hardware. And that is the way an application should
be.
>Independent of the hardware.
>
>    First lets look at the worst case.  Power drops in the middle of a
write
>operation.
>With this all bets are off.  This is why we have a recovery log.  Before an
update
>is made, the record is read and saved in the recovery log.  The updates are
made to
>the record and also saved as the update in the recovery log.  Now we go and
write
>to the data base. Then the data is read back and checked.  If it went onto
the disk
>correctly we now clear the entry in the recovery log. To insure that the
data is
>safe these are the operations we put the hardware through:
>        1: Writer the recovery log:  Save the update request
>        2:  Read the data base:  Save the old data
>        3: Write the recovery log:  Save the merge of the update with the
data in
>the data block.  AIX writes 4k blocks of data.  Most updates only change
part of
>the data.
>        4: Write the data to the disk.
>        5: Purge the disks cache.  AIX will first search the devices cache
to see
>if the data is still there. This will save time.  We want to know what is
on the
>disk so we force AIX to go to the disk to get the data.
>        6: Read the recovery log:  Compare the log to the data read.
>        7: Write the recovery log.  Clear the entry.
>                From the above you can see we did 7 I/O operations to
change one
>byte.   This how we protect our data from errors during update.  It's been
a few
>years since I did this type of work so I'm sorry if I left something out.
>
>        Mirroring protects our data while it is static, just sitting out
there.
>
>AIX Mirroring:
>        To do proper AIX mirroring you should have three or more drives.
This is
>because AIX requires 51% of the drives to be online to do a write.  For the
ADSM
>recovery log we use only two drives.  If ADSM can not write to the recovery
log it
>will abort without changing the database.  AIX mirroring allows you to
write to one
>drive and at the same time read from the other.  This gives you the best
>performance, but requires twice the disk space.
>
>RAID5 Mirroring:
>        Raid 5 protects the data by writing part of the data on each disk
and
>adding ECC (Error Correcting Codes) to each part.  If one disk fails the
ECC on the
>other disks will allow a full recovery of  the missing data.  The ECC only
adds
>about 20% to the total disk space required. AIX mirroring adds 100%.  There
are no
>free lunches! (Unless you work in a Casino)  Because you are using all the
disks
>together you can not read and write at the same time.  Because the disk
motors run
>at slightly different speeds the read/write operations will start and end
at
>different times for each disk.  This makes for slower disk operations.
RAID 5 is
>slow, but cheap.
>
>I hope this adds some understanding to the mirroring question.
>
>ex-hardware engineer
>ex-diagnostic programmer
>ex-AIX device driver programmer
>ex-IBMer (32 years)
>Never could spell
>Bill Dias
>AIX Systems Administrator
>Brown & Sharpe
>
>Richard Sims wrote:
>
>> >I've always thought that hardware mirroring is more efficient than
software
>> >mirroring.
>> >The only reason that I have heard on this list to use ADSM mirroring is
far
>> >fetched at best so I would suggest using hardware mirroring if possible.
>>
>> We've EXHAUSTIVELY discussed this in the past.  And the value of ADSM
mirroring
>> has been demonstrated: it's not far-fetched.
>>
>> Newcomers:  Please have a look in the bountiful information available
>> at www.adsm.org for general topics, which you can bet have been addressed
>> over the years.
>>
>>    Richard Sims, BU
<Prev in Thread] Current Thread [Next in Thread>