Explain to me whether I need a battery backup for my RAID card

ZippyDan

Platinum Member
Sep 28, 2001
2,141
1
81
So I'm getting a bunch of Dell PERC 6i/R RAID cards for some Dell servers.

Now I know that some RAID cards come with a little battery (BBU or TBBU) to save the write cache in case there is a power loss to the machine. I understand that there can be corruption if the machine shuts down during a write.

My question is then, why do RAID cards without batteries even exist if this is such a dangerous problem? Why would anyone use a RAID card without a battery? There seem to be a lot of them around, so what exactly is the application when a RAID card without a battery is fine?

None of these servers will be general file servers... they are email servers, domain servers, inventory/ordering server (so that is a small database), etc.

Last question, I've heard that SSD drives change the necessity for a RAID battery since they have larger write caches?
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
Without battery they use Write Trough. With battery they use Write Back.

Write Back is obviously much faster. Also if you force Write Back without battery backup. Its not just corruption during a write. You can essentially lose 100s of MB of unwritten data.

SSDs dont really have write caches. They are simply faster.

The cache is mainly to hide the awful seektimes and poor IOPS on HDs. Usually 10000-14000us vs 50-85us for SSD. And 200-500 IOPS vs 50000-80000 on SSDs.

In short, the battery is a MUST.
 
Last edited:

ZippyDan

Platinum Member
Sep 28, 2001
2,141
1
81
Without battery they use Write Trough. With battery they use Write Back.

Write Back is obviously much faster. Also if you force Write Back without battery backup. Its not just corruption during a write. You can essentially lose 100s of MB of unwritten data.

SSDs dont really have write caches. They are simply faster.

The cache is mainly to hide the awful seektimes and poor IOPS on HDs. Usually 10000-14000us vs 50-85us for SSD. And 200-500 IOPS vs 50000-80000 on SSDs.

In short, the battery is a MUST.

Ok this is what I wanted to get to. Assuming that performance is not a huge factor for me, what is the worst that could happen if I don't have a battery? As you said when you mentioned write-through and write-back, it seems that the RAID card is smart enough to operate without a battery by being more conservative about its writing. Why is a battery a MUST? And if it is a MUST, why do so many battery-less server-class RAID cards exist?

Also you didn't really address whether a BBU is necessary with SSD drives?

Another question that just jumped into my mind (and a quick google doesn't seem to answer): do NAS boxes like Synology or QNAP tend to use write-through or write-back? (do they have a BBU?)
 
Last edited:

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
If you dont force Write Back without battery. It will default to Write Through without battery or with a failed battery. Then the worst part is just much lower performance.

Most got battery. Those that dont today usually got a mix of capacitors and flash backed cache.
 

Micrornd

Golden Member
Mar 2, 2013
1,341
221
106
The question of battery or no battery really depends on how critical the data is.

Without the battery most RAID cards can still be forced into write-back and this is fine for home/general usage when used with a UPS, since it will allow a controlled shutdown and the data will be written to disk in that time period.

Where the battery comes into play with critical data is when you have a total (as in PS supply/MB/CPU, etc.) failure and there is no chance of the data being written to disk.
The battery will allow the cache memory on the RAID card to retain the data in RAID card cache memory (at the time of power loss) for several days, allowing the card and drive/array to be brought back on line, either in a new or repaired machine and the retained cache data to be written the drive or array.

Do pay attention to the underlined above.
Nothing is perfect, if you are doing a copy, that's one thing, BUT if you are doing a cut and paste and the data/file is larger than the RAID card cache memory and power is lost, the data/file is lost.

That's why the norm is to copy and verify.
If you were running a program at the time, the same similar scenario holds true, so hopefully you were saving often.

Li-ion batteries are available from both the OEMs and third parties.
For a Perc 6i they run about $50 new, just be sure you get or have the connecting cable too. :biggrin:

*********************************************************
OK, all that being said, you need to be sure of what you are getting to know if this even matters -

PERC6i are for Medium to large Servers which need high performance and redundancy. It supports RAID0,1,10,5,50,6 and 60. It has cache of 256MB or 512MB and a BBU (battery backup Unit) and directly supports 8 disks via 2 x 4 port cables

PERC6i/R is an entry level "RAID Controller" that supports RAID0 and 1.
It has no BBU capability and supports 8 disks via 2 x 4 port cables.

So if you are getting PERC6/iR's, battery backup doesn't enter the picture :whiste:

And a link to all Dell RAID cards - http://www.dell.com/learn/us/en/04/campaigns/dell-raid-controllers
 

ZippyDan

Platinum Member
Sep 28, 2001
2,141
1
81
The question of battery or no battery really depends on how critical the data is.

Without the battery most RAID cards can still be forced into write-back and this is fine for home/general usage when used with a UPS, since it will allow a controlled shutdown and the data will be written to disk in that time period.

Where the battery comes into play with critical data is when you have a total (as in PS supply/MB/CPU, etc.) failure and there is no chance of the data being written to disk.
The battery will allow the cache memory on the RAID card to retain the data in RAID card cache memory (at the time of power loss) for several days, allowing the card and drive/array to be brought back on line, either in a new or repaired machine and the retained cache data to be written the drive or array.

Do pay attention to the underlined above.
Nothing is perfect, if you are doing a copy, that's one thing, BUT if you are doing a cut and paste and the data/file is larger than the RAID card cache memory and power is lost, the data/file is lost.

That's why the norm is to copy and verify.
If you were running a program at the time, the same similar scenario holds true, so hopefully you were saving often.

Li-ion batteries are available from both the OEMs and third parties.
For a Perc 6i they run about $50 new, just be sure you get or have the connecting cable too. :biggrin:

*********************************************************
OK, all that being said, you need to be sure of what you are getting to know if this even matters -

PERC6i are for Medium to large Servers which need high performance and redundancy. It supports RAID0,1,10,5,50,6 and 60. It has cache of 256MB or 512MB and a BBU (battery backup Unit) and directly supports 8 disks via 2 x 4 port cables

PERC6i/R is an entry level "RAID Controller" that supports RAID0 and 1.
It has no BBU capability and supports 8 disks via 2 x 4 port cables.

So if you are getting PERC6/iR's, battery backup doesn't enter the picture :whiste:

And a link to all Dell RAID cards - http://www.dell.com/learn/us/en/04/campaigns/dell-raid-controllers

Great information. So I'm assuming that with the PERC6/iR, write back is not even an option, or at least not a good idea?

I do have hefty UPS's for my servers, but I would assume that that only makes things better if they servers are AWARE of the UPS status and can perform a graceful shutdown before the battery dies.

Finally, I still don't have an answer on whether using write back is advised when using an SSD RAID?
 

gea

Senior member
Aug 3, 2014
241
17
81
You must care about two problems.

One is the controller or disk cache where a sudden crash/powerloss can lead to a dataloss/ inconsistent filesystem. This is a raid independent problem and can be solved when you disable any disk or controller write caching. This is slow but safe.

The main problem with a raid-controller is the write hole problem with non CopyOnWrite filesystems like ext4 or ntfs. In such a case it can happen that a crash during a write may result not only in an inconsistent filesystem but also an inconsistent raid for example in a raid-1 where one disk is updated wheras the second disk is not or a raid-5/6 where a stripe is written partly to some disks - without any chance to detect the problem without a checksum filesystem like ZFS. In such cases a controller cache + BBU can help to reduce the problem.

If you want to avoid these problems completely, you need a CopyOnWrite filesystem where a write is done completely or not done at all - no partly updates (btrfs, ReFS, WAFL/NetApp, ZFS) + software raid where the OS has full control over disks + realtime checksums to detect any problems + a SSD logdevice (ZIL in case of ZFS) with a supercap. Much better (faster and safer) than any hardware raid with cache and BBU. Mostly you use NetApp or ZFS as btrfs and ReFS are not yet ready for serious use.
 
Last edited:

corkyg

Elite Member | Peripherals
Super Moderator
Mar 4, 2000
27,370
239
106
Does all this local stuff apply if the entire computer is backed up automatically by a good UPS?
 

gea

Senior member
Aug 3, 2014
241
17
81
A good UPS is always a good idea for critical data but it does not protect against a crash due a software or hardware problem and it adds an additional risk for a power failure.

You can solve this with a redundand PSU where one line is connected to AC.