RAID level for Redo

Hi,
My storage admin created RAID10 and RAID5 for database, I would like to know which RAID Level is best for keeping the REDO logs. Can someone tell me what's best for REDO?
Thanks

hello,
In case of using redo logs then you need performance and availability.that is why i put this comparison for you.
I suggest you use raid 5 because it is faster in the write operations and more recoverable.*
<pre class="jive-pre">Note: RAID 5 disks are primarily used in the processes that require transactions. Relational databases are among the other fields that run very well under a RAID 5 storage scheme*</pre>
RAID 5 vs RAID10
Data Loss and Data Recovery
Let us start off by having RAID 5 explained. In RAID 5, the data backup of any one of the disks is created. If there are 5 disks, in the storage system, then 4 of the disks will be used for storing the data and one of the disks will be used for keeping the backup of any one of the hard disks. If one of the disks in the array fails, then the data can be recovered, but in the event of a second disk failure, the recovery is not possible. RAID 10 on the other hand is a combination of RAID 0 and RAID 1. In a RAID 10 storage scheme, an even number of disks is required. Each disk array has a disk array, which is a mirrored set of the former. In RAID 10, data recovery of all but one disk can be performed. In the case of a disk failure, all the remaining disks can be used effectively without any impact on the storage scheme.
Performance
The RAID 5 performance in the read operations is quite appreciated, though its write operation is quite slow, as compared to RAID 10. RAID 10 is thus used for systems which require high write performance. Hence, it is very obvious, RAID 10 is not used for systems like heavy databases, which require high speed write performance.
Redundancy
The RAID 10 arrays are more data redundant than the RAID 5 arrays. This makes RAID 10 an ideal option for the cases where high data redundancy is required.
Architectural Flexibility
RAID 10 provides more architectural flexibility, as compared to RAID 5. The amount of free space left is also minimized, if you use a RAID 10 data storage scheme.
Controller Requirement
RAID 5 demands a high end card for the data storage performance. If the purpose of the RAID 5 controller is being solved by the operating system, then it will result in the slowing down of the performance of the computer. In case of a RAID 10 controller, any hardware controller can be used.
Applications
RAID 10 finds a wide variety of applications. Systems with RAID 0, RAID 1 or RAID 5 storage schemes are often replaced with a RAID 10 storage scheme. They are mainly used for medium sized databases. RAID 5 disks are primarily used in the processes that require transactions. Relational databases are among the other fields that run very well under a RAID 5 storage scheme.
With this, I complete the RAID 5 vs RAID 10 comparison. This comparison, I hope, will help you in deciding the right storage scheme, that can suit your purpose.
kind regards
Mohamed

Similar Messages

The best RAID level for video editing that has some form of redundancy?

I've been asked to help find the best solution for importing and editing large amounts of HDV (25 Mbit/s) video. However, those whom I am helping also want a level of redundancy that will allow a single drive to fail and their data to be preserved. So what I'm trying to figure out is the best RAID level (or levels) for their need. I'm fairly certain that either 0+1 or 10 is what I'm looking for, however the I/O performance differences between 0+1 and 10 aren't quite clear to me. If someone could explain that to me I would appreciate it. Additionally, if someone knows a better level than either 0+1 or 10 for the needs I described, please don't hesitate to let me know.

The difference between RAID 10 and RAID 0+1 is how the array is created.
RAID 0+1 creates a RAID from multiple RAID 0 arrays that are mirrored.
You can tolerate any number of drive failures in any one side of the mirror (that side of the mirror goes offline as soon as any one disk fails, so it doesn't matter how many other disks fail in the same array), but one drive failure on both parts of the mirror will trash your data.
RAID 10 creates a stripe of multiple RAID 1 mirrors.
In this setup you create RAID 1 mirror sets and stripe them together.
In this setup you can tolerate one disk failure in each element of the stripe.
For example if you have 6 drives you might create three mirrors of two drives each, and each mirror is then striped.
You can lose one drive in each mirror, but if you lose two drives in any mirror set, you're out of luck.
In both cases, though, you're not going to get the best usage out of the array since mirroring has a 50% overhead - meaning you're only going to get 50% of your total disk capacity as usable space.
With the XServe RAID, the RAID 5 performance is very good - good enough for your 25MB/sec throughput so I'd go with RAID 5 arrays, not RAID 0 or 1 on the XServe RAID itself.
Then, depending on your space requirements, you can either stripe or mirror them together for a RAID 50 or 51 array. In this way you gain the data redundancy of RAID 5 with better disk utilization than 10 or 0+1.
RAID 50 will give you the best performance (and the most usable space), RAID 51 will give you the best redundancy.
At the end of the day it's up to you to decide which format to use based on performance and usable space requirements.

Raid levels for sql server files

Hi,
I have a doubt on raid levels.
What RAID level should be used for SQL Data files and why?
What RAID level should be used for SQL LOG files and why?
What RAID level should be used for Tempdb files and why?
Thank you.

SQL Log File - The I/O is sequential in nature but only if you have a single database on the raid group of disks. If you have 2 or more database each doing its sequential pattern, you result having random I/O on the raid group.
SQL Data files - yes they are random in nature and could be write intensive but consider that SQL Server writes synchronously only in the log file, the writes on the data files are made in background by lazy writer and check point so it is not mandatory
to have RAID10, I will always choose RAID 5. The write penality of the RAID5 does not affect performances.
One last consideration: modern storage do "wide striping" i.e. they spread I/O operation on a pool of raid group. Every read/write operation first hit the cache of the storage (battery protected RAM) if the cache has not the blocks needed by the
rad operation or has not the amount of space to satisfy the write operation, the I/O is demanded to the disks...
Work with your storage admin if you have one and ask him to do cache partitioning and reserve some only for SQL Server log, you will be amazed by the gain in performances...

RAID Levels for SAP BW

I need to know if different raid levels are recommended for SAP BW than for other SAP applications such as R/3. We are having load performance issues and RAID is one area we are looking at as a potential improvement area.

Hi Jeff,
due to the fact, that SAP has included BI (BW) into NetWeaver, there could no different recommendations for BI and ERP (R/3). I would recommend, that you first should try to find out if you really have an I/O problem.
If you want to find out if you have I/O problems, use TA ST04->"detail analysis" to get the I/O times of the database. On an Oracle system you can get the I/O times on volume level down to the data file level. You'll find the I/O times in milli seconds. You should check if your I/O times are more or less the same on all of your volumes. If not may be you should move some files to disks with lower I/O times.
Normally I/O times should below 10 ms. I/O performance is not good or bad, only you can decide if the performance is good enough or not. As a rule of thumb I would say, that the performance is the better, the more disks you are using. But on current I/O subsystems you have plenty of parameters to vary. Number of disks, stripe size, striping on OS or subsystem level, RAID level, cache size and file system parameters (logging, buffering ...).
As a hint, we run systems on different I/O subsystems and we have I/O times between 12 ms down to 0.8 ms. Needless to say, that the costs of the different systems varies a lot. The 0.8 ms could not be achieved by using only disks, you can only achieve it be using huge caches independent of the RAID level. RAID 10 (1+0) is probably better if you have a lot of random I/O, from a price performance perspective RAID 5 could be better because you have more active disks (more disk space) for the same amount of money.
The bottom line is, you have to decide yourself if your I/O is good enough.

Best RAID configuration for storing Datafiles and Redo log files

Database version:10gR2
OS version: Solaris
Whis is the best RAID level for storing Datafiles and Redo log files?

Oracle recommends SAME - Stripe And Mirror Everything.
In the RAC Starter Kit documentation, they specifically recommend not using RAID5 for things like voting disk and so on.
SAN vendors otoh claims that their RAID5 implementations are as fast as RAID10. They do have these massive memory caches...
But I would rather err on the safer side. I usually insist on RAID10 - and for those databases that I do not have a vested interest in (other than as a DBA), and owners, developers and management accept RAID5, I put the lead pipe away and do not insist on having it my way. :-)

Storage raid levels are required for ASM?

Storage raid levels are required for ASM? Can we have both storage and asm level of raid configuration on the server?

Handle:      user530956
Status Level:      Newbie
Registered:      Sep 15, 2006
Total Posts:      59
Total Questions:      26 (26 unresolved)
So many questions, without ANY answers.
http://forums.oracle.com/forums/ann.jspa?annID=718
Can we have both storage and asm level of raid configuration on the server?YES

ASM Disk Group RAID Levels

This is the scenario that I am currently working on. Just need some input on whether it is feasible or not.
We have a 2 node RAC running Oracle 10.2.0.3 on AIX 5L. Database size is ~2TB. The database mostly performs OLTP but also stores some historical data.
There are two main applications using the database - one performs high reads with some small updates & inserts, while the other is very write intensive but does some reads as well.
Currently there are three disk groups one for the tablespaces (dg_data), another for system/sysaux/undo tablespaces (dg_system) and another for archived logs & redo log copies (dg_flash) - all using external redundancy. ASM best practises recommend no more than 2 disk groups. It also recommends disk groups with disks of similar characteristics including raid levels. However, the dg_data disk group has both RAID 5 and RAID 1+0 disks which house tablespaces for both applications. Seeing that the applications have different requirements (heavy reads vs heavy writes) does it make sense to create a separate disk group with 2 different RAID levels or would using RAID 5 in dg_data satisfy both requirements?

I am attempting to generate some statistics on the ASM Disks I/O activity before implementing the disk group separation in order have some metrics for comparison purposes. Enterprise Manager Grid Control displays the performance of disk groups and individual disks by showing the Disk Group I/O Cumulative Statistics. When comparing the results with the asmiostat output I am unable to correlate the results. I know that the asmiostat queries the v$asm_disk_stat view. Where does EM GC pull it's information from?
For example, I run the following query on the ASM instance:
SQL> select group_number,disk_number,total_mb,free_mb,name,path,reads,writes,read_time,write_time,bytes_read,bytes_written
2 from v$asm_disk_stat
3 where group_number=
4 (select group_number from v$asm_diskgroup
5* where name = 'DG_FLASH')
GROUP_NUMBER DISK_NUMBER TOTAL_MB FREE_MB NAME PATH READS WRITES READ_TIME WRITE_TIME BYTES_READ BYTES_WRITTEN
1 0 8671 8432 DG_FLASH_0000 /dev/asm2 14379476 10338479 149205.75 19633.64 290,136,450,560.00 7.2165E+10
1 1 8671 8431 DG_FLASH_0001 /dev/asm3 11470508 10278698 184597.5 19313.54 249,769,027,584.00 9.2911E+10
1 2 8671 8432 DG_FLASH_0002 /dev/asm4 17274529 8743188 178547.56 38342.52 339,439,240,192.00 6.7165E+10
The output from the same period on Grid Control is below
MEMBER DISKS AVG RESPTIME AVG THROUGHPUT TOT I/O TOT RDS TOT WRTS RDERRS WRTERRS
DG_FLASH_0000 5.58 2.58 33179503 21949607 11,229,896 0 0
DG_FLASH_0001 8.26 1.83 25752100 13131695 12,620,405 0 0
DG_FLASH_0002 8.11 1.86 28269693 18798823 9,470,870 0 0
The statistics in the query are lower than those in the EM GC report. I also tried querying the fixed views (x$) but the results were even more confusing.
What is the best method for comparing and gathering statistics on ASM activity?

RAID level, ASM

Hi all,
from performance perspective: Which RAID levels are recommended to store OCR/Voting disks , Redo logs, Control files, datafiles in an 11gR2 RAC using ASM?
- OCR/Voting disks with NORNAL redundancy ASM level.
- Redo logs, Control files, datafiles, temp files EXTERNAL redundancy ASM level.
(we will use Redhat Enterprise Linux 5.5)
Thank you,
Diego

Diego wrote:
from performance perspective: Which RAID levels are recommended to store OCR/Voting disks , Redo logs, Control files, datafiles in an 11gR2 RAC using ASM?A 2-way mirror with a quorum disk is needed for the OCR and voting disk. This mean at minimum 3 disks.
For the database - that depends entirely on what redundancy you need for the database layer.
ASM automatically stripes across the disks for a diskgroup. You can then choose to mirror that in addition. Or you can use external mirroring (on the storage server/SAN). Or you can use RAID10 on the SAN and then use single stiped and mirrored LUNs per diskgroup. Or you can use multiple such LUNs per diskgroup, which means ASM will stripe the striped set. Not a real issue and covered in an Oracle support note.
As ASM does not support RAID5/6/7, you can use that on the physical storage layer and then simply stripe in ASM.
Or you can use 2 storage servers and use ASM to stripe the disks (in a diskgroup) per server. Then mirror these across storage server boundaries, thus introducing redundancy at physical storage server layer.
Lots of possibilities and combinations. The best being whatever meets your requirements the best.
Also keep in mind that the fabric layer also need to be redundant. No use of having 2 storage servers for example and mirroring across these for redundancy, when connectivity to the storage servers are via a single non-redundant fabric layer switch. Or wiring a dual port HBA/HCA into the same switch (cable failure covered, but loose the switch and loose all connectivity to the fabric layer),

Avoid RAID-5 on Redo Logs- WHY?

Hello all,
I read that, one should avoid using RAID Level 5 on Redo Log Files. Can anyone help me make understand this concept?
Why it is recommended so?
Also, it will be a great help if you can provide me articles explaining the various levels of RAID and their effect/significance on database.
Thanks in advance.
Himanshu

In general, RAID usually impacts write operations more than read operation.
This is specially true where parity need to be calculated (RAID 3, RAID 5, etc).
Online or archived redo log files can be put on RAID 1 devices.
You should not use RAID 5. 'TEMP' tablespace data files should also go on
RAID1 instead of RAID5 as well. The reason for this is that streamed
write performance of distributed parity (RAID5) isn't as good as that of
simple mirroring (RAID1).
Jaffar

Hitachi Array Configuration (Cache - RAID levels ? on Oracle9i/Solaris9)

I shall install Oracle9i Release2 on a Sun Enterprise 4500 with 12 processors using a Hitachi 9200 array with a two-controller, 512M each, and 10 73GB drives). IT recommended and decided no fail-over and/or clustering.
1) Does anyone have any previous experience in configuring the cache arrays with specific recommendations. We plan to set the redo logs on that cache as a leading-edge mechanism from that vendor. Are there any known drawbacks.
2) Hitachi suggests that their technology is fast enough to handle RAID 5. What RAID levels do you recommend for permanent, in particular for indexes, and also for temporary datafiles? Oracle listed recommendations have been RAID 0+1 on datafiles; and RAID 1 on redo log files, and archivelog files.
3) Has anyone implemented successfuly RAID 5 on Hitachi before ? The database server will primarily be an MTS with some hybrid EIS features for primarily overnight processing and two consumer groups are associated.

Great work vango44!
Here are some RAID performance statistics I gathered while testing RAID on my system. The testing software was Winbench 99. The hard drives tested were new Seagate ST380013AS drives, formatted NTFS. Winbench was running on a third drive that is not included in the tests and should not affect the results.
The drives were reformatted between tests and chkdsk'ed to try and keep things "apples to apples".
No hardware or software changes other than the RAID setup/connections were made between tests.
Higher numbers mean better performance.
I also ran the same tests on the newish WD Raptor 10K drives:
I couldn't stand all the noise the Raptors made, so I returned them.
On my motherboard:
SATA 1 & 2 = Intel RAID controller
SATA 3 & 4 = Promise RAID controller
If the test title does not include "RAID", then it was a single drive test.
Unfortunately, I don't have a spreadsheet version of the above stats. Otherwise I'd create nice bar charts for us and it's would be easier to deduce performance.
Perhaps some kind reader will OCR the pictures, put them into Excel, and make some nice bar charts for us?
Hope the info helps.

RAID Levels

Hello, I know that there are many different ways that a database can be set up with regards to RAID levels but what i have found is that you get told what disks you have available and then try to work around it.
Our infrastructure guys want to move my 2 of my databases from the storage they are on. ISCSI drives on the SAN to free up some space.
They have offered me
2 sets of local mirrored pairs of disks on the server and potentially some RAID 5 on the SAN for datafiles and RAID 10 for logs.
Now,
The local disks are big so it is a shame to not use the space.
Would the following work ok do you think?
Option 1
Mirrored Pair 1 : (C: Drive (Windows and Oracle Binaries) + Redo Logs
Mirrored Pair 2 ( Data Files and Archive Logs)
Can we have Redo Logs on the same drive as the O/S without potential performance problems? There isn't massive activity on these databases. Generally only 0.6 log switches per hour (Usually there is a peak of activity for a few hours, perhaps 6 log switches each hour and then only 1 or 2 for the remaining hours in the day. 50MB size redo logs.
I presume having archive logs on the same as the data files storage with the low amount of log switching shouldn't be a performance problem.
Option 2 (Not keen as these are disks with a few hundred GB but would only be having a few hundred MB used
Mirrored Pair 1 : (C: Drive (Windows and Oracle Binaries) + Archive Logs
Mirrored Pair 2 : Redo Logs
SAN RAID 5 - Data Files
The info i have found is that Archive/Redo Logs should be on RAID 1.
Datafiles should be on RAID 10, if not RAID 5 or RAID 1
Any advice as to what to do based on what i have available to me?
Thanks

I couldn't tell ya, since I only have a few things on windows and don't like it. But I'll make these observations:
Redo is Oracle's achilles heel. Lose it and you fail. You want it on multiple devices in addition to mirroring. Slowest device determines one bottleneck (with some possible modern exceptions).
Modern raid-5 is ok, except when it's not. The latter includes when a disk fails. Disks fail more often than mtbf implies.
Check your apps load profile. For my situation, undo is the most used data file. YMMV.
As has been said, more spindles is more better. But the real determinant is data flow. If you put everything through one controller, you might have a bottleneck. If there is a net involved, something else might be on the net. You need to consider everything from where Oracle requests data to what happens on the spinning rust, or cycling electrons on an SSD, as the case may be. So if you have Windows randomly deciding to do stuff on a disk, and that interferes with redo, you can have mysterious slowdowns. I couldn't tell you whether your setup would have this as an issue. I can tell you sometimes a bored Oracle on Windows will take some time to wake up when you poke at it.
Redo and archiving redo have sequential read and write characteristics. Data files have random read and write characteristics. Best not to mix the two if possible.
You can worry about this until the cows come home, but in the end, you still have to empirically determine how your app actually works. This can change over time, sometimes radically with very little provocation. So try to leave yourself some wiggle room if whatever you decide on changes - and be ready if your organization suddenly can let some more hardware drop from above.
I've heard good things about SLOB (Silly Little Oracle Benchmark), worth googling. Good luck.

Advice on RAID Sets, Volume Sets, and RAID Levels of the Volume Sets using an Areca Controller

I have read through a lot of information on disk usage, storage rules for an editing rig, users inquiries/member responses in this forum and I thank each and every one of you – especially Harm.
In building my new workstation, I purchased five (5) WD 1T, 7k, 64M SATAIII hard drives and an Areca RAID card, ARC-1880ix-16-4G, which I plan to use primarily as my media/data disk array. The workstation will use a 128GB SATAIII SSD as the OS/program drive and I will transfer two (2) WD Raptor/10k SATA 70GB drives from my current system for pagefile/scratch/render use. I tentatively plan on using a mobo SATAIII port for the SSD and mobo SATA ports with a software RAID (level 0) for the 10k Raptors.
In reading the Areca Instruction manual, I am now considering exactly how I should configure the 5 physical 1TB drives in terms of RAID Level(s), Volume Sets, and RAID Sets. I must admit that I like the opportunity of allowing for a Dedicated Hot Spare as I am generally distrustful of the MTBF data that drive vendors tout and have the bad experience in the past of losing data from a mal-configured RAID array and a single drive hardware failure (admittedly, my fault!).
In line with the logic that one doesn’t want to perform disk reading while trying to write at the same time (or vice-versa), I am thinking the approach above should work OK in using the mobo disk interface and both software and external hardware RAID controllers without having to create separate RAID level configurations within a Volume Set or further dividing up the physical drives into separate RAID sets. I know in forum messages that Harm noted that he had 17 drives and I could envision a benefit to having separate RAID sets in that situation, but I am not at that point yet.
To some degree I think it might be best to just create one RAID Level on one Volume Set on one RAID Set, but want to solicit thoughts from veteran controller users on their workflows/thoughts in these regards.
Anyone care to share thoughts/perspectives? Thanks
Bill

Thanks for the speedy feedback Harm - I appreciate it.
I was thinking RAID level 3 as well.
Of course, it's always something! I purchased the Caviar Blacks by mistake - which are non-TLER. I will work with EggHead to return the ones I purchased and replace them with RE4 versions as I'm not thrilled about the possibility of the controller declaring the volume/disks degraded unnecessarily and although I have the DOS utility WDTLER where one is supposed to be able to enable/disable TLER on WD drives - I suspect WD is way beyond that now anyway with current builds.
I agree with you about just testing the performance of the options for the raptors - on the mobo and then on the controller. When I benchmark them I'll post the results in case others are curious.
Thanks again....off to EggHead!

Raid Level 5+6 under 10.5.6 ?

we couldn't use our SCSI RAID Level 6 under 10.52/10.53/10.54/10.5/5
heavy CPU load after a while and crash of the AFP
now the RAID is for sale !
Murphy's low, seems that Apple has fixed some kinds of problems with RAID
http://support.apple.com/kb/HT3192
any body has already Experience with the 10.5.6 and the use of RAID Level 5 or Level 6 on a Server?
any information would help
thanks

No version of Mac OS X supports RAID5 or 6. Theses RAID levels are always implemented in hardware and therefore not subject to the whims of the OS. In reality the RAID is shrouded by the controller and OS has no idea of the underlying array format and there there is nothing to fix in the OS.
I think the pertinent part of your post, though is the 'SCSI' element. I expect that ant problems you have relate to the SCSI card you're using and not the OS. Since there are no specific SCSI fixes in 10.5.6 I doubt anything has changed (but Apple are notorious for not detailing all the low-level changes they make, so you might be lucky.
You'd have a better chance, though, of looking at the SCSI card you're using (what make/model is it?) and checking the drivers provided by the vendor. Most SCSI problems in my experience have come down to card and drivers, not necessarily the OS.

X4500 RAID Configuration for best performance for video storage

Hello all:
Our company is working with a local university to deploy IP video security cameras. The university has an X4500 Thumper that they would like to use for the storage of the video archives. The video management software (VMS) will run on an Intel based server with Windows 2003 server as the OS and connect to the Thumper via iSCSI. The VMS manages the permissions, schedules and other features of the cameras and records all video on the local drives until the scheduled archive time. When the archive time occurs, the VMS transfers the video to the Thumper for long term storage. It is our understanding that when using iSCSI and Windows OS there is a 2TB limit for the storage space - so we will divide the pool into several 2TB segments.
The question is: Given this configuration, what RAID level (0, 1, Z or Z2) will provide the highest level of data protection without comprimising performance to a level that would be noticable? We are not writing the video directly to the Thumper, we are transferring it from the drives of the Windows server to the Thumper, and we need that transfer to be very fast - since the VMS stops recording during the archiving and restarts when complete, creating down time for the cameras.
Any advice would be appreciated.

I'd put as many disks as possible into a RAID 5 (striping) set. This will provide the highest level of performance, with the ability to sustain a single disk failure.
With striping, some data is written to all the disks in the stripe set. So, if you have 10 disks in the set, then instead of writing data to a single disk, which is slow, 1/10th of the data is written to each disk simultaneously, which is very fast. In effect, the more disks you write to, the faster the operation completes.

RAID level

Hi,
I'm building a three node RAC OLTP system, which RAID level is better for performance? Is there any document available?
Thanks,
GK

Depends on your application. If there was always one right answer there would be only one flavor of RAID.
The general rule for Oracle is to do what ASM does: SAME
Stripe And Mirror Everything.
Also check this out: www.baarf.com
Any organization with Juan Loaiza, Cary Millsap, and Mogens Nørgaard as members is worth paying attention to.

RAID level for Redo

Similar Messages

Maybe you are looking for