Posts Tagged ‘JBOD’

SQL Server 2014 Fast Track

November 25, 2014

Over the past several months the major hardware vendors have been hard at work developing detailed documentation around how to build an optimal data warehouse using SQL Server 2014. My friend and MSFT APS (PDW) TSP James Serra (you get a prize for knowing what all that means) has already blogged about it here. But I would like to add a few of my own observations as well.

HP:

HP DL380 Gen8 and HP LE PCIe Workload Accelerator 28TB/45 TB Data Warehouse Fast Track Reference Architecture

HP DL580 Gen 8 and HP LE PCIe Workload Accelerator 90TB Microsoft SQL Server Data Warehouse Fast Track Reference Architecture

Generally speaking, I have been impressed over that last few years with HP dedication in the DW space. This is based on having worked closely with them in past with FastTrack and PDW/APS. I like that they are making use of JBOD storage which also used in the APS (aka PDW)

 

HP sql server 2014

 

JBOD is a much more efficient approach and still preforms the same as the classic storage array/enclosure and storage controllers/processors but usually at a lower cost.

I am also happy to see that they stayed with the strategy of spreading database files from multiple file groups over the LUNs/mount points.

hp sql server 2014 2

This is that same basic setup that I cover in my DW videos: Data Warehouse 5 Parallelism.

I was much less impressed with the EMC FTDW solution it smells of EMC trying to force fit it’s VNX 5600 storage platform into the DW world. Most of the document feels like a VNX OLTP sales pitch that stresses things that are not major considerations for a DW solution.

Example:

p.9 “It allows the storage system to provide flash-drive class performance to data with a high locality of reference, which increases IOPS without placing all the data onto flash drives”. This sounds great for an OLTP solution (random reads/writes) but not for DW (sequential reads/writes).

“Multicore Fully Automated Storage Tiering (FAST) Cache reduces the load on the hard drive (HDD) by absorbing I/O bursts forms applications…” Again sounds good for OLTP but not applicable for DW. How do I/O bursts work for a DW? How would auto-tiering handle large analytical/aggregate queries i.e. last year sales aggregated by quarter? Where is the I/O hot spot there?

However, if EMC is you corporate standard you may have little choice in the matter and so choosing their FastTrack solution maybe your best option.

In summary

Not much has changed in the SQL Server 2014 FastTrack. Other than Clustered ColumStore Indexes are updatable and give world class performance. I think that JBOD, much to some SAN vendors dismay, is the future of at least DW if not OLTP storage.