John Andersen wrote:
> On 9/11/2014 3:26 PM, Linda Walsh wrote:
>> Compare RAID 5, 6, 50, 60 against the same using linux
>> mdraid setups. Check your throughput as well as cpu usage.
>> I think you'll find the mdraid version notably more taxing on
>> your cpu.
>
> Nope.
>
> Even on relative old hardware mdraid never makes it to the
> top of top no matter how busy the system is. CPU load is
> trivial. Disappearingly so.
---
That wasn't my experience with RAID5 or RAID6 (didn't try 50/60 at
the time). RAID6 was near unusable.
> Its pretty clear you are just mouthing the platitudes of raid card
> vendors.
---
Yes, actual testing showing differences is what I believe
in. With the same disks, RAID5 w/md5raid topped out at under 300MB/s
writes, ~600MB reads. HW raid at the time was 500-600MB/s writes
and ~1000MB/s reads.
Run some tests, that's what we did. Even in the days
> of 486 processors as servers, raid-cards never did pay, and we
> sent the expensive ($4000 as I recall) back.
---
Dual processor +1G cache RAID card have seen listed for
a hair under 500. A new Xeon from Intel: 1200-1800. Now some
of those figures are a few years old, but $4000???... maybe back
15-20 years ago?
I do remember the md-driver being less demanding in what drives
it allowed. It would accept any drive -- with speeds varying by as much as
20% from slowest to fastest in the desktop HD segment.
The LSI controller failed 11 out of 14 desktop drives for being
out of spec. (desktop drives had mistakenly been sent instead of server).
Variance of drives accepted as 'GOOD' by LSI controller: <1%.
> With today's newer multi-core processors you simply can not
> make a business case for raid cards over MDRaid.
---
Spending $500 to NOT trash system cache and get a 2x I/O
boost that constantly monitors flags and rebuilds seems like it
would be worth it. That's also for an extra eight 6Gb ports you wouldn't
have without some sort of HW outlay.
I havn't benched the md-drivers lately, when I did they were not
yet multi-cpu capable (so all checks and rebuilds happened in a single
core. I know that has changed, which would tend to hide cpu usage
with it getting done about 10x faster on a 10-core Xeon.
I also had the experience of having my md-raid5 corrupted
twice after a system crash before I decided it wasn't quite
what I wanted in reliability.
--
To unsubscribe, e-mail: opensuse+unsubscribe(a)opensuse.org
To contact the owner, e-mail: opensuse+owner(a)opensuse.org