[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: measuring RAID arrays performances



On 5/26/2013 5:40 AM, Jean-Marc wrote:

> OK.  Thank's for your mail and sorry to make you loosing your time reading mine.

Don't apologize.  I'm not wasting my time here.  Quite the opposite.
Hopefully I'm teaching not only you, but others, that trying to
optimized performance of a -2 disk- array is a waste of -your- time.

> So, apparently, I have to read a lot of doc'.  Because launching a massive read on the same devices did not give the same result when I took a look at <iostat -k 3> output:

Docs won't fill in the gaps in your knowledge here.  Neither will
running bonnie++, iozone, using 'dd', etc.  They simply do not reflect
real world use of a 2 disk RAID1.  RAID10,f2 on two disks is simply an
optimization of RAID1 and it's not going to double your throughput.
Whether it is of benefit to your depends entirely on your workload(s).
And with only 2 disks and using the installer, it's pretty clear that
your workload is not 100% single threaded streaming read, which is where
f2 yields the bulk of its benefit.  In real world use for a desktop or
most small server workloads, random read/write performance is important,
not sequential read.  This is why I said stick with vanilla RAID1.

The numbers you posted, which I'm snipping, are irrelevant.
<snip>

> Is iostat output wrong ?
> Are these reading speed and volume wrong ?
> Do I have to rely on this or not ?

You tell me.  What is your workload?  If it's other than mostly
streaming read you'll not benefit much, if any, from the f2 layout.

>> You really need to read and comprehend much more about md, or simply
>> listen to an expert:
>>
>> For two disks, use RAID1.  PERIOD.  End of story.  Screwing with
>> anything else is a massive waste of your time.  If you have 3 or more
>> disks, then you can worry about layouts.
> 
> Could you also send this advice to Debian people ?
> 
> Because the layout current in use inside the official Wheezy Debian installer for a 2 disks system is RAID10,f2.

Defaults are selected for many reasons.  They are not always the best
nor always sane.  One need look no further than the decision to -not-
include non free firmware, making it difficult or impossible, depending
upon experience, for users to get GPUs and 802.11 devices in laptops
working properly.

> It is may be time to open a bug for this.

You'll want to go through the installer defaults for all disk counts
before filing such a bug report.  There are likely others that many
people would disagree with.  If the 3 disk default is RAID5 then ~30 of
users will disagree saying it should be RAID10,f2 for better performance
and redundancy at the cost of space.  If it is RAID10,f2 then ~70
percent will disagree saying the extra space is more important than
redundancy and performance.  Advanced users won't care because they'd
-never- use the installer to create their arrays.

With only 2 drives, the overall performance difference between RAID1 and
RAID10,f2 is nearly statistically even, with real world workloads.  So
to reiterate my previous point, it is a waste of -your- time attempting
to figure out which one is slightly faster.  It is your time to waste,
so go ahead if you like.  I'm simply trying to save you some that can be
put to better use elsewhere.

-- 
Stan


Reply to: