Michael Sheets, the IBM performance guru for System z, is a straight shooter and passionate about the integrity of performance documents. When he sends me something, I always read it. Yesterday he sent me an IBM Technical Brief “IBM System z® and System Storage DS8000: Accelerating the SAP® Deposits Management Workload with Solid State Drives, March 9 2009”.
There were many parts in this paper, and I have extracted the pieces that are of interest to the storage community. Mike and his team took the SAP® Deposits Management Workload (part of Banking Services for SAP 6.0), and investigated the use of Flash Drives on performance. They looked at two components of the Deposits Management workload, the postings workload, and the account balancing workload. Both are I/O intensive workloads.
Table 1 shows the data I extracted from the technical brief. The two storage environments are very different. The first is a traditional configuration with 896 hard disk drives (HDD) in 3 IBM 8300 storage arrays. The second is a single array with 192 drives, 96 HDD and 96 Solid State Drives (SDD).
As you can see from table 1, the SSD reduces the electrical power and cooling consumption to about 24% of the HDD and the space is reduced by two thirds. The approximate price is about 20% higher.
The SAP posting workload is a traditional OLTP benchmark with a very high number of DB2 database call and fewer I/Os to the logs. The impact of the SSD configuration is significant, with 22% additional throughput enabled, the response time improved by 1.7 seconds.
The SAP account balancing workload is very different. It is a batch type of run, in which the database logs are nearly 50% of the IO, which is characterized by high bandwidth and high write ratio. The performance impact of SSD is very different from the OLTP transaction, and external throughput is reduced by 1%.
In a previous professional alert I said that there were two main benefits from flash drives:
- The ability to perform hundreds of times more I/O that traditional disks and replace large numbers of disks that are I/O constrained.
- The ability to increase system throughput by reducing I/O response times
These measurements illustrate both benefits. In I/O intensive workloads the number of drives was reduced from 896 to 192. In the OLTP workload system throughput was improves by 22%. That is a saving of 6,000 MIPS, at a street cost of well over $6 million dollars for hardware and software. Table 1 shows that the additional cost of the flash storage configuration was less that $500,000.
This measurement is born out in the real world. In 2008 a financial company was not meeting its SEC compliance obligations to process the batch by the start of the next daily cycle. The solution before flash drives were introduced was to add an additional 1,000 MIPS of mainframe processing power, which would have increased the total system power by over 25%. That would have cost more than a million dollars in hardware and software, compared with the implementation of 16 EMC flash drives.
The measurements also show that workload is very important in understanding the impact that SSDs will have on performance. The OLTP workload experienced a 22% improvement in throughput, and the balancing workload showed a 1% reduction in throughput with this particular storage configuration. The key will be to look at the workload that determines the server capacity required by the applications.
Action Item: As the relative price of SSDs comes down from its current 10:1 figure to a 3:1 figure over the next two to three years, the value of applying flash drives to storage configurations will increase. These measurements and case studies show that flash drives will be used to manage I/Os, and SATA drives will store bytes. For application architects, flash drives can enable higher numbers of DB calls and higher DB locking rates, helping to improve overall application functionality.
Storage architects and executives should be positioning themselves for a radical change in storage architecture, especially as storage vendors introduce arrays that can automatically and optimally place data at a sub-volume level.
Footnotes: Note 1 - Source: IBM System z® and System Storage DS8000: Accelerating the SAP® Deposits Management Workload With Solid State Drives, 3/9/2009 Downloaded on 3/12/2009 from http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP101442