Where to put 3 M1015 and an ARC1200 controller on the X9SCM-F?

[dikkiedirk]

I recently added a 3rd M1015 controller to my main server (further specs are in my sig). It seemed to work rather well with good speed and parity checks running at an average speed of 87 MB/s. I now have converted all disks in the array to XFS and removed a 2 TB disk from the array. I did this by setting a new config and had to rebuild parity. This parity build took more than twice as long as the parity check at a speed of 40 MB/s from start to finish. It seemed this speed was capped to 40 MB/s. Can this be because I don't have a parity disk but a RAID0 parity volume that sits on a ARC1200 controller? Or is it the fact that the ARC1200 is on the outermost PCIe slot of the X9SCM-F motherboard. Or is it some configuration error?

Can anyone share his thoughts on this?

 

Would it matter which of the x4 slots to use for the ARC1200? One is directly on the CPU and the other on the PCH.

Would that matter for the 3rd M1015?

[WeeboTech]

Do you have write caching enabled on the arc-1200?

 

 

There could be contention on the bus also, that's allot of drives!

[SSD]

I just did a rebuild of a data disk in my array. It started out at 125 MB/sec slowing down to about 60 MB/sec at the end (I was rebuilding a 3TB drive and all disks were >= 3TB).

 

My parity build speed was much much slower writing to my raid0 ARC 1200. If I remember it started in the 85 MB/sec range and got slower from there.

 

Rethinking if the ARC is doing me any good.

[StevenD]

I pulled my ARC-1200.  I wasnt seeing any benefit from it at all.

[WeeboTech]

The benefit is small for sequential writes. Unless you have the write cache enabled.

For me, parity create speed was greater then parity check speed due to the write caching.

 

 

With high random writes, the arc-1200 provided 'me' a big benefit in the architecture of the machine at that time.  X7SBE.

 

However with today's drive speeds reaching in the 200MB/s range along with the sysctl vm.highmem_is_dirtyable=1 parameter, it doesn't show that much of a benefit.

[dikkiedirk]

 

However with today's drive speeds reaching in the 200MB/s range along with the sysctl vm.highmem_is_dirtyable=1 parameter, it doesn't show that much of a benefit.

 

What does this parameter do? And where do I set it. I run V6 beta10.

[SSD]

One thing I like about the ARC is the ability to combine the capacities of two drives to create parity. So 2 3T drives can handle up to 6TB. This prevents the need to buy two larger disks in order to benefit from the higher capacity.

[dikkiedirk]

Well the parity has been build now, took a bit long, but parity checks only take 11-12 hours at an average of 88 MB/s and peaking at 140 MB/s. The ARC1200 will remain in service a little longer but might get retired when I move to a 6 TB or larger parity disk.

[WeeboTech]

One thing I like about the ARC is the ability to combine the capacities of two drives to create parity. So 2 3T drives can handle up to 6TB. This prevents the need to buy two larger disks in order to benefit from the higher capacity.

 

 

and for me, it was having the combined safe mode. i.e. RAID0 parity and RAID1 cache. That RAID1 cache saved my butt big time!

[dikkiedirk]

Sure an ARC1200 has advantages, but on the X9SCM-F, in combination with 3 M1015s, You have to make choices. Put an M1015 or ARC1200 on the slowest PCI-E slot? By slowest I mean the one that isn't directly on the CPU, but on the chipsets PCH. There will be a performance hit nevertheless. But will a M1015 with 8 disks suffer more than an ARC1200 with 2 disks in RAID0, that's the main question in my specific case.