MSA1000 large array

[Ralph Hebb]

Does anyone have information regarding best practices or benchmarks for sizing arrays on an MSA1000, specifically one that will hold 1+0 LUNS? I have 20x146GB drives and am considering creating a single array so that my LUNS all have a large number of spindles to work with. Is this sane, or should I consider creating smaller arrays; if so, what would be the benefit? Thanks in advance.

[Jefferson Humber · Accepted solution]

Ralph,

Some might see having one large array as putting all your eggs in one basket ? (Personally I have 2 arrays on my MSA).

To create a 20xHDD array you will of course need to split the array over a secondary storage enclosure to accomodate.... something happens to the enclosure you'd loose the whole array.

But of course having 20 spindles would give better performance. Is performance or resilience more important to you ?

As long as your not using RAID5 I don't see too many issues in having 20 spindles in the array, just make sure you have built in resilience i.e 2 x PSU's e.t.c

Jeff

[raadek]

I would say that spanning RAID 1+0 array across two enclosures rather improves the reliability: as long your one 'mirror' is within one enclosure & the other 'mirror' in the second one, the array will survive a failure of any entire HDD enclosure.

[Jefferson Humber]

I'm not sure you have control over the placing of the mirrors.

[Ralph Hebb]

Jeff, I believe you are correct that the MSA1000 does not allow a specification of mirror location. Is there any information on how the MSA1000 allocates mirrors?

I do have 2 shelves, so it sounds like my plan is sound so long as I consider all points of failure.

[Basil Vizgin]

Multiple LUNs is not recommended at one big array if you produce heavy write utilization of a LUN and simultaneous read access of another LUN on the same array group:
http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?lang=en&cc=us&taskId=110&prodSeriesId=415598&prodTypeId=12169&objectID=c00622089

[Ralph Hebb]

Basil, is that limitation considered a defect by HP that will be corrected in later firmware, or is it inherent to the archetecture of the MSA1000? Is there a maximum number of LUNs reccommended, or is this behavior seen anytime there is more than one LUN per array?