DSM-G600, DNS-3xx and NSA-220 Hack Forum

markchicobaby wrote:

Ouch, this guy is an writer for Atomic, a PC publication that focuses on high performance hardware.  He lost his RAID5 array, and has no backup.

Well worth reading.  If you need convincing that RAID != backup, this article will do it.

http://www.atomicmpc.com.au/article.asp … e=20080317

RAID is all about performance (0) and availability (1). Nothing more and nothing less. Unfortunately too many people think backup when they hear "redundant," so I hear this conversation too often: "Is this data backed up somewhere?" "Yeah, I have RAID 1."

fordem wrote:

Nothing more and nothing less.

Actually there's quite a bit more to RAID than just 0 & 1 - try 0, 1, 3 or 4 (depending on who's counting) 5, 6, - and there are the combination levels such as 0+1 & 1+0, which are both known as RAID10 even though they are not the same and likewise 50 & 60.

And in case it isn't obvious - some of those are about both performance AND availability - the RAID implementation needs to be selected based on the application, and you can include cost as a factor to be considered - it is not an either/or situation.

Thanks for letting me know there are other RAID levels; I can see you know a lot of technical details. That's very impressive. Still, it boils down to availability (simplest implementation: 1) and performance (simplest implementation: 0). RAID 0 and 1, therefore, sums up the major benefits of a RAID setup; 0 for performance because of striping; 1 for availability because of redundancy. All other levels (including nested levels) enhance these simplest implementations but do not offer anything other than availability AND performance. And I did say "and" the first time too.

Nothing more and nothing less.

Last edited by blahsome (2008-03-17 18:00:50)

jayas wrote:

Hello,

The term backup refers to making of additional copies of data that can be used to restore the original in the event loss of data occurs.

In the case of RAID1, the purpose of mirroring is exactly that.  Thus IMHO it is not incorrect to refer to RAID1 as a backup method.  Just because the "restore" operation takes place automatically it does not follow that mirroring is somehow not for purposes of backup.

From a practical perspective, the user may backup data from one disk to another disk.  Usually this is done periodically, say once a week, and again usually the backed up drive is often kept off-line.   In RAID1, this backup process occurs far more frequently, and the backup drive is kept on-line to enable automatic restore on disk (hardware or software) failure.

My 0.3 cents worth

Jaya

Agreed. Strictly and technically speaking, you are correct.

Practically speaking, considering RAID1 as backup is often dangerous. One has to understand that RAID1 offers no protection against high-level file corruption/deletion/overwrites, some of the most common occurrences that need a previous version of the files affected.

IMO, it's rather pointless discussing every little technical detail regarding RAID and backup. The end users won't care. We need to think in terms of the common data loss scenarios and see how well they are covered. RAID can be reliable (data is always there, unless you do something to it) and fast (data is retrieved quickly), but it's far from a complete backup solution needed by end users. In that regard, it's always simpler to think that RAID is not backup, or at least not a form of backup for some of the most common data loss scenarios.

Last edited by blahsome (2008-03-17 18:21:43)

jayas wrote:

Hello,

The term backup refers to making of additional copies of data that can be used to restore the original in the event loss of data occurs.

In the case of RAID1, the purpose of mirroring is exactly that.  Thus IMHO it is not incorrect to refer to RAID1 as a backup method.  Just because the "restore" operation takes place automatically it does not follow that mirroring is somehow not for purposes of backup.

From a practical perspective, the user may backup data from one disk to another disk.  Usually this is done periodically, say once a week, and again usually the backed up drive is often kept off-line.   In RAID1, this backup process occurs far more frequently, and the backup drive is kept on-line to enable automatic restore on disk (hardware or software) failure.

My 0.3 cents worth

Jaya

Not quite correct.

With RAID1 there is no backup and no restore, automatic or otherwise, what exists is a single virtual drive, "concealing" the two physical drives, it is mapped as a single drive, with each physical drive being written to simultaneously as if there were only a single drive, reads are treated differently and either drive may be read from.

Should either physical drive fail in this scenario, the user will be none the wiser and will continue to access his/her data from the remanining physical drive, the complexities being completely hidden by either the RAID controller or the operating system.

Now - consider this scenario (which I have seen used in another low cost linux based NAS, and called mirroring there) - your NAS has two hard drives, one is mapped and used by the attached systems as storage, the second is unmapped and holds backups made from the first drive, using rsync.  This is quite similar to the scenario in your third paragraph - except of course for that automatic restore, which is not going to happen automatically in the case of a drive failure, until the drive is physically replaced.

If the first drive fails, none of the attached systems have access to the data stored on the second drive until, either the failed drive is replaced and the data restored, or the drive mappings changed to point to the remaining drive, the "currency" of the data is dependent on what changes were made to the data on the first drive, since the last rsync backup.

If the second drive fails, none of the attached systems are in any way affected.

Please note - I have deliberately omitted the necessary alerting & management parts of the process in the interest of brevity.

I don't disagree with your definition of backup at all - what I disagree with is the procedure that you outline.

The terms backup and restore imply the creation of a copy of the data from the original in a different location (or on a different medium) and then a reversal of that process, with the data being copied from the backup location or medium to the original location.

This is not how RAID works - yes - with RAID1 two copies of the data will exist, but, it is not an original and a backup copy of the original, consider it as two originals, made simultaneously on different physical medium (disks) , and should one of those disks fail, there is no "restore" - the user simply continues to access the original data from it's original location.

Hi Fordem,

fordem wrote:

This is not how RAID works - yes - with RAID1 two copies of the data will exist, but, it is not an original and a backup copy of the original, consider it as two originals, made simultaneously on different physical medium (disks) , and should one of those disks fail, there is no "restore" - the user simply continues to access the original data from it's original location.

Your notion of two originals is interesting.  Reminds me of XEROX commercial where the salesman challenged a potential customer to identify which of of the two documents he produced was the original.   After a while, the customer looked at the salesman and said "Ah ...  you are trying to trick me.  Both of these are originals!".

Jokes aside I understand what you are saying.  It is a matter of semantics.  If you look at "backup" as a functional requirement, then RAID1 is one implementation that will meet such a requirement under certain circumstances.

Regards,

Jaya

Last edited by jayas (2008-03-18 09:00:14)

RAID1 allows for quick recovery in case of a single disk failure and this is what I am after when using RAID1 in my office computer/server/dns-323. When a single disk fails I can still run the system in degraded mode and wait for the RMA to arrive. On the day of a disk failure I would immediately query the running disk for SMART stats to see if it is also likely to fail. Do not overestimate the SMART data, Google (as a company) claims that only 30% of all catastrophic disk failures can be predicted by monitoring the SMART info.
RAID 1 does not substitute for a backup, for a number of reasons:
1. You cannot restore mistakenly deleted files from RAID 1 as compared to backup
2. In case of theft/fire/lighting strike there is no recovery path (the same applies for backups kept under the same roof!)
3. In rare scenarios of double disk failure (as per above) you are done

RAID1 on a proper HW controller should in theory double the read transfer speed of the disks on large files by alternating data streams from each disk, but this is a dream not to come through on most software RAID, fake RAID (ich9r, etc) and cheap hardware controllers.

I personally do not endorse RAID 5 as it makes little sense now when 1TB disks are readily available for a reasonable price and the heat/noise generated by the array is only justifiable in rack mounted installations.

RAID1+0, RAID 10 etc. provides great service when there is a high demand on data stream bandwidth such as high end database applications, video editing, etc.

Overall RAID and backup a completely different technologies with a different scope of use and deployment.

And the golden rule- there is never enough backups. Do not be shy to overdo them.

fordem wrote:

Nothing more and nothing less.

Actually there's quite a bit more to RAID than just 0 & 1 - try 0, 1, 3 or 4 (depending on who's counting) 5, 6, - and there are the combination levels such as 0+1 & 1+0, which are both known as RAID10 even though they are not the same and likewise 50 & 60.

And in case it isn't obvious - some of those are about both performance AND availability - the RAID implementation needs to be selected based on the application, and you can include cost as a factor to be considered - it is not an either/or situation.

You are a stickler for semantics and one-upmanship like I've never seen before.

Hi Grimham,

Grimham wrote:

RAID 1 and up = Fault Tolerance - not backup.

One of the purposes of backup is fault tolerance.

Jaya

Hi Mig,

mig wrote:

I seems this topic is a source of frequent confusion...

Wikipedia http://en.wikipedia.org/wiki/Backup  wrote:

Due to a considerable overlap in technology, backups and backup systems are frequently confused with archives and fault-tolerant systems. Backups differ from archives in the sense that archives are the primary copy of data and backups are a secondary copy of data. Backup systems differ from fault-tolerant systems in the sense that backup systems assume that a fault will cause a data loss event and fault-tolerant systems assume a fault will not.

While the Wikipedia is not an authoritative source of of information it normally provides a good insight into usage.  In the context of this thread however it seems add to the confusion by differentiating archives from backups when in the *nix environment, tape archive or tar format is used for backup purposes.

Even the Wikipedia itself here http://en.wikipedia.org/wiki/Tarball states

Initially developed as a raw format, used for tape backup and other sequential access devices for backup purposes, ...

Of course tar (tape archive) format is not the only form of backup but it is one that widely used even today.  Have a look here: http://www.usenet-forums.com/linux-netw … -help.html

Jaya

Last edited by jayas (2008-03-19 10:58:02)

I have to agree with fordem on this one.  There IS much confusion going about it.  Just as a summary to everything, I'm going to very simply lay it out below.  (I do mean very simply, I understand the complexities behind RAIDS, but am going to put it in layman's terms.)

All semantics aside:

1) RAID systems are not back-ups.  As stated before, they are used for speed and/or redundancy.  A redundant RAID array is to keep a system running after [typically] a hardware failure, usually of a hard drive.  In RAID 1, for example, both copies of the data are "live" and can be read, written, and executed.  If a user deletes something, it is gone from both.

2) A back-up is an independant copy of data that is set at a fixed point in time.  It's purpose is to recover after a failure of any kind to a fixed point in time.  This failure could be hardware, software, or even a user that deletes something.  It is kept in an offline state and is used only to restore data.  It is only written to once and is not edited (though it may be entirely overwritten), at least not in good practice.  Often multiple back-ups are kept at different increments in the event the failure is captured in one of the back-up copies, there is another that can be restored.

3) An archive is typically original data that is no longer used and is being stored away (often offline) in the event that it may be used sometime in the future.

Hopefully this gives a good enough layman's description and can put this issue to bed.

Last edited by bq041 (2008-03-19 20:57:33)

Hi Bg041

bq041 wrote:

I have to agree with fordem on this one.  There IS much confusion going about it.  Just as a summary to everything, I'm going to very simply lay it out below.  (I do mean very simply, I understand the complexities behind RAIDS, but am going to put it in layman's terms.)

Attempting to rehash a debate in the name of simplification has its dangers -- it tends to be loaded to support a particular view

IIRC the original question was that given DNS-323 has RAID1 feature that allows one to recover from drive failures, can this feature be used in lieu of traditional backup and recover methods for purposes of defending against drive failures.  My short answer is yes, provided one understands disk mirroring and how it is similar and different to traditional backups techniques involving one or more of snapshots, checkpoints, journaling, incremental, etc ...

AIUI the follow up question related to whether or not the off-site backups can be implemented by swapping drives.  My short answer is also yes, provided one understands one has to shutdown the DNS-323, and also ensure that replacement drive is 'blank' before DNS-323 is restarted and go through the GUI interface and initialise it.  I predict the next release of firmware will support hot-swapping which will remove the above restriction.

Was fun engaging in semantics but I think we had enough sort of.

Jaya

PS:  I will add that by and large I do not disagree with the alternative views expressed.  I just think they distract too much from the question(s) raised by OPs.