Re: Raid1 replaced with raid10?

[Peter Rabbitson <rabbit@rabbit.us>]

Neil Brown wrote:
> On Monday May 7, rabbit@rabbit.us wrote:
>> Neil Brown wrote:
>>> On Friday May 4, davidsen@tmr.com wrote:
>>>> Peter Rabbitson wrote:
>>>>> Hi,
>>>>>
>>>>> I asked this question back in march but received no answers, so here it
>>>>> goes again. Is it safe to replace raid1 with raid10 where the amount of
>>>>> disks is equal to the amount of far/near/offset copies? I understand it
>>>>> has the downside of not being a bit-by-bit mirror of a plain filesystem.
>>>>> Are there any other caveats?
>>>>>   
>>> To answer the original question, I assume you mean "replace" as in
>>> "backup, create new array, then restore".
>>> You will get different performance characteristics.  Whether they
>>> better suit your needs or not will depend largely on your needs.
>> Hi Neil,
>> Yes I meant take an existing 2 drive raid1 array (non bootable data) and
>> put a raid10 array in its place. All my testing indicates that I get the
>> same write performance but nearly double the read speed (due to
>> interleaving I guess). It seemed to good to be true, thus I am asking
>> the question. Could you elaborate on your last sentence? Are there
>> downsides I could not think of? Thank you!
> 
> I would have thought that you need "far" or "offset" to improve read
> performance, and they tend to hurt write performance (though I haven't
> really measured "offset" much).
> 
> What layout are you using?
> 

Correct, I am using 'far' layout. The interleaving of the 'offset'
layout does not work too good for sequential reads, but far really
shines. Yes write performance is hurt by about 10%. Compared to 190%
gain in reads I can live with it.