Re: Mirror Oracle-Redologfiles on Oracle-Level even when on Raid-1 or EMC-Systems ?

If the EMC array is truly 'state of the art', it should allow you to do the following:

1. partition some of the physical disks such into a large mirrored RAID-1 partition and a small non-mirrored partition to hold the Oracle logs (which Oracle apparently wants to mirror manually)
2. give the small non-mirrored logical log volumes priority space in the Symmetrix write-back cache (which is battery-backed for stability, as I recall): this will greatly expedite your log writes with no sacrifice in safety (even in the highly unlikely event that the cache develops a bad board, the copy several miles away will still be valid), and at the same time minimize any performance interactions with the other (non-log) data that shares the same physical disks (every once in a while a large chunk will get written back to disk, but far less frequently than the Oracle log 'forces' occur).

Being expert neither in Oracle software nor EMC hardware, the above is just a guess at what may be feasible. Good luck, and let us know how it turns out.

• bill

Harald Wakonig <wakonig_at_compuserve.com> wrote in message news:3878C431.F7D60B66_at_compuserve.com...
> Hi,
>
> Within another thread on comp.databases.oracle.server ws1_at_netcom.com
> wrote:
>
> WS> Even if place on RAID-1 arrays, the redo logs should
> WS>still be mirrored at the WS> Oracle level.
> WS> If the log writer process determines that it does not
> WS> know whether the contents of a particular redo log are valid,
> WS> it will mark that redo log as "STALE" in the v$log table.
> WS> If this redo log is the only copy, then it cannot be archived,
> WS> and will cause a database halt.
>
> * How often does this really happen, that the log writer marks a redo
> log as "STALE" ?
>
> * What is the typical reason that this happens ?
>
> * Could such a redolog-file marked as "STALE" be used during an instance
> recovery ?
>
> The problem of an additionally mirroring on Oracle-Level are the
> COSTS...
>
> Please don't argument that you can buy a 4-GB disk for a PC for a few
> dollars...
>
> The following Oracle-related questions are not only valid for
> EMC2-storage arrays, but are valid for ALL (expensive) state-of-the-art
> storage arrays from other vendors:
>
> Facts:
> * We use EMC2 storage array , Raid-1 for ALL disks (each disk is
> mirrored in the local EMC2 storage-array)
>
> * We use BCV's (3rd mirror, which can be separated after short
> database-shutdown , allowing a cold-backup to tape although the database
> is again running on the other 2 mirrors)
>
> * Additionally we use the SRDF-Option of EMC2 (Symmetrix Remote Data
> Facility), thus mirroring every write-call to a second EMC2-storage
> array in a second data center (some miles away...), where it is again
> mirrored within that EMC-box.
>
> * Currently I have 4 Redo-groups with 100-MB redo file size .
>
> For performance reasons its suggested to keep the redolog-files on one
> reserved disk...
>
> * Our EMC2 storage array contains only disks with size of 18-GB
> (No of the Oracle-DBA's - was asked by the department which bought the
> storage array)
>
> The consequences are: for 4x100 = 400 MB
>
> 2 x 18-GB disks (Raid-1 mirrored) in the local EMC-box
> 1 x 18 GB disks (local 3rd mirror, "BCV")
>
> 2 x 18 GB disks in the remote EMC-box in the second data center
> 1 x 18 GB disk "BCV" (third mirror) in the remote EMC-box
> ==============
> 6 x 18 GB = 108 GB disk space for the redolog files
>
> If I will do an additional mirroring at Oracle-Level, I just would need
> another 6 disks (108 GB disk space)...
>
> Some tuning experts suggest to stripe Oracle-Redolog-Files for faster
> write rate ... that would again double the required disk space ....
>
> But even if I would have only one local Raid-1 mirror I would need 2x18
> GB = 36 GB disk space for the redolog-files (still not striping...), and
> the complete net data volume is about 50 GB ...
>
> If you know the costs of disk space in such state-of-the-art storage
> arrays you know that those costs are some factors beyond the price of an
> 4-GB disk at a PC discounter... (and the costs the Oracle department or
> the project gets internally charged from the UNIX department operating
> the disks contain additional dollars for the administration, usually a
> certain amount per Gigabyte...
>
> As many Oracle customers operate databases on expensive storage arrays,
> how is this topic typically solved ?
>
> Idea 1: Do no mirroring for the Disks containing Oracle-Redolog-Files
> ================================================
> Problem: The UNIX department operating the EMC-box will make HIGH
> additional internal charges to the Oracle department because of
> additional work ..... (Our internal ISO-9000 - checklist have mandatory
> mirroring, we would need to add additionally exemptions for the case
> that mirroring is done by an application, this additional rules must
> pass the QA-department ...)
>
> Idea 2: Put Oracle-Redolog-Files on local disks, which are cheaper..
> ==============================================
> * No chance if you want to mirror everything to a second data center ...
>
> * No chance if you use BCV's or any other "third mirror" for fast
> cold-backups
>
> Idea 3a: Buy some 4-GB spindles instead of 18-GB spindles
> ========================================
> * Are such configurations (18- and 4-GB spindles supported by the
> storage array vendors ?
> * How many Oracle DBA's are asked by the UNIX department BEFORE the
> contract with the storage array vendor is signed ?
>
> Idea 3b: Buy an additional Storage Array ONLY for all
> Oracle-redolog-files of all your databases
> ==================================================================
> That's of course only for sites with many databases ...
>
> * We have currently 3 EMC-Arrays which are mirrored with SRDF to a
> second Data center (6 EMC-Boxes)
>
> For easy operating I could imagine, that we by an additional set (1 Box
> for each data center), but only with small (4 GB spindles), and the
> default settings, that in this system NO local mirroring is activated,
> only the SRDF mirroring into the second data center.
>
> The costs for the 4th system can be realistic, because there is always a
> need in more disk space, and by moving all redologs to the new system
> the same space becomes available on the existing 3 storage arrays.
>
> BUT each UNIX server will need a SECOND host adapter, because it will be
> connected now to two different EMC boxes (on most UNIX Servers we use
> only ONE host adapter, on some we use two). And that are definitely
> additional costs.
>
> * Idea 4: Share the redolog disks with other databases
> ======================================
> * As we operate many databases using the same EMC2 box it might be
> possible, but:
> - the databases reside on different UNIX servers, and each UNIX server
> has one or two dedicated EMC-host-adapters (Channel adapters). I don't
> know if it is possible, to have on one physical EMC-spindel (which is
> linked to exactly one EMC-Disk-Adapter (DA)) more file systems, which
> are dedicated to UNIX servers using DIFFERENT EMC-Channel-Adapters...
>
> * This will not improve the performance .....
>
> * Idea 5 : Put other Oracle-Files with LOW usage on that disk
> ==========================================
> What would you suggest ?
> * Oracle-Binaries ?
> * temporary tablespace (I have usually less then 5 disk sorts an hour
> .., I just need the temporary tablespace during "analyze" ..
>
> - But that are only 3 GB... in that case I would have 3,4 GB used out of
> 18 GB
>
> * Some projects use huge staging areas (keeping several old
> database-exports....) - If the export is run, than there is usually no
> redo-activity, and its possible to live with the "read-export-file" and
> "write redolog" in case of the very seldom imports ...
>
> Thank you for your comments or hints to further information (white
> papers...)
>
> Harald
Received on Sun Jan 09 2000 - 16:39:23 CST