Sample Cost Benefits Analysis

Cache Memory Upgrade on

Mainframe Disk Controller

(3990-002 to 3990-003)

Pat Brown

Director of MIS

January 29, 1990

Table of Contents

I. The Problem

II. The Recommended Solution

III. Technical Feasibility

IV. Net Present Value Analysis of Acquisition Options

V. Tangible and Intangible Costs

VI. Tangible and Intangible Benefits

VII. Future Considerations

Appendix A. Supporting Graphs and Charts

Appendix B. What is Cache Memory and How Does It Help?

I. The Problem

When the Order Entry and Distribution System was implemented in production, the degradation of CICS response time on all on-line transactions was reported by the users to be greater than acceptable. Within MIS, several tasks to reduce response time were accomplished including:

n increasing the number of buffers for the database

n balancing the files over the disk storage devices

n improving the application programs utilization of keys to the database data

The steps taken have reduced the response time, however, the reduction is not sufficient to enable the order department to process orders and ship within the time available. Response time must be reduced further.

II. The Recommended Solution

It is the recommendation that our company acquire a 32 megabyte cache memory upgrade tot he 3990-002 disk controller. We further recommend that option four(4) be the means for the acquisition. Option four is to buy the controller(currently being leased), buy the cache memory upgrade and finance through the XXX Credit Corporation.

III. Technical Feasibility

A “DASD Audit” (audit of our disk storage utilization) was conducted to determine if either cache memory or additional disk storage would help with the response time problem. The study shows a cache memory upgrade at the 32 megabyte level will aid in bringing response time down.

Two time periods were studied, 4 p.m. and 10 p.m. The two time periods were selected as worst case times for response. The users indicated that while average response times were inconvenient, the worst case response times were debilitating.

The study was conducted on every volume of disk storage. Five conditions were tested:

1) current response time

2) 32 Mb cache response time

3) 32 Mb cache with software/hardware to increase channel speed from 3 Mb/second to 4.5 Mb/second

4) 64 Mb cache response time

5) 64 Mb cache with software/hardware to increase channel speed as above.

The result is as shown on the accompanying graphs (Appendix A). For the 4 p.m. time frame, the disk volume with the worst access time (60.6 milliseconds) is DB005 (Exhibit 1). The average access time for all volumes is 27.6 milliseconds (Exhibit 2) The improvement for DB005 when utilizing the 32 Mb cache memory is a reduction of 79% to 13 milliseconds. The worst case volume at 32 Mb cache is DB0011 which proved to be a poor cache candidate. The average reduction across all volumes is 12 milliseconds or 43%. Adding software/hardware to increase channel speed did not reduce response time significantly. Utilizing 64 Mb cache reduced the average access time from 15.6 milliseconds to 11.4 milliseconds or 27% reduction. The additional software/hardware at the 64 Mb cache level also had negligible effect on access time. The 10 p.m. time frame graph (Exhibit 3) shows similar results. From a technology perspective, adding the 32 Mb upgrade to the disk controller (3990-002) significantly recudes access time and makes good sense (Exhibit 4).

A benefit not covered in the DASD Audit is the ‘fast write’ feature of cache memory. This feature reduces access time for disk updates from the 30 millisecond range to about 4 milliseconds. The customer services department functions, where heavy updating typically occurs, may benefit as much from ‘fast write’ as from other cache memory features.

A large reduction in disk access time does not result in an equivalent end-user response time improvement. Other factors such as CPU utilization and exclusive control conflicts also affect overall transaction throughput.

Currently CPU utilization is running at 90% on first shift and 78% on second shift. By improving disk access times, we will experience an increase in CPU utilization. It is reasonable to expect an increase of 5% on each shift.

Exclusive control conflicts occur when a task needs to update a record that another task has under its control for update purposes. Speeding up disk access will help this situation to some extent, but additional application program tuning will more significantly reduce or eliminate the problem.

There are no measurement or modeling tools that will tell us exactly what the addition of cache memory will do to the overall throughput picture. However, it is the consensus of the MIS technical support and management staff that adding cache is the next logical step in improving the mainframe’s performance. We have a reasonable expectation that we will see 10% to 20% improvements. Putting this number in terms of user satisfaction, currently we process approximately 152,000 transactions per day. A twenty percent improvement would result in the potential to process 182,400 transactions per day. Also batch processing currently running in ten hours (i.e. GO1030 - batch order allocation) would run in eight hours.

At this time the order entry on-line allocation is not implemented in production. The on-line allocation is such a resource hungry process that having it in production literally halted processing. The programming staff have applied their intellect to the problem and have drastically improved the algorithm for the process. While still burdensome, we will be trying it in production in next week. Because much of the resource it requires is tied to input/output processing, it will be particularly helped by adding cache memory to the disk controller.

IV. Net Present Value Analysis of Acquisition Options

The options considered for acquiring the cache memory upgrade are:

1) purchase the cache upgrade ($124,500),continue the lease on the disk controller (3990-002) until termination date, then buy the controller.

2) extend the lease on the disk controller and lease the cache upgrade

3) purchase the cache upgrade ($124,500), exercise the purchase option on the controller lease now ($101,000)

4) purchase the controller and cache upgrade utilizing XXX Credit Corporation for financing.

In every case, MIS is comfortable with utilizing the 3990 disk controller for the next five years. The 3990 disk controller is current technology and there is nothing on the horizon that will displace it. Upgrades may be required if the work loads change, but the basic box will be current technology for at least five years.

Also in every case, the maintenance for the 3990 will be increase from $316 per month to $684 per month (utilizing our xxxxxxxxx service agreement which discounts at 17% on service).

Several assumptions are made in the net present value analysis. One is that we know today what the fair market value of the controller will be in 1992 at the end of our lease and furthermore, we know today what the fair market value will be for the upgraded controller will be in 1993. Nothing could be further from the truth. We have values in 1993 for the upgraded controller that range form $84,000 to $175,595. The range for the controller in 1992 is $55,440 to $85,500. These two ranges skew the analysis. If we take the low end, the results are completely opposite than if we take the high end. Knowing this, we have selected values in between. We have selected the value of $75,000 to purchase the controller 1992 for Option 1 and further more, we have selected $112,750 for the sale price of the upgraded controller in 1993. The break-even value on the controller is between $80,000 and $85,000 which we feel is too low. The net present value is calculated using these values. Looking at the result, Option 2 is a lease option and it is much higher. Options 1, 3, and 4 are close.

In Option 1, we are purchasing an upgrade for a lease box. At the end of the lease, we buy the box for fair market value. I find this an untenable position. Option 3 and Option 4 are close to equal for net present value. Technically, they are both ‘buy’ decisions so leave MIS with a box in place which probably has a longer useful life than marketing trends would indicate.

In looking at Option 3 and 4, XXXX financing is using an interest rate lower than our current cost of capital, so I recommend Option 4.

NET PRESENT VALUE ANALYSIS FOR ACQUISITION OF CACHE MEMORY UPGRADE FOR DISK CONTROLLER

Option 1: Buy Cache Memory, Continue lease on DISK Controller, Buy Controller at end of lease for FMV (est. $85,500)

	1990 Feb-Dec	1991 Jan-Dec	1992 Jan-Dec	1993 Jan-Mar
Current Lease on Controller	-26,785	-29,220	0	0
Lease Buyout (Comdisdo)	0	0	-75,000	0
Cache Memory Upgrade purchase	-124,5000	0	0	0
Sell Controller	0	0	0	112,750
SUBTOTAL	-151,285	-29,220	-75,000	112,750
Tax Credit(lease)	10,312	11,250	0	0
Depreciation (Cache)	24,900	39,840	23,904	14,342
Depreciation (Controller)	0	0	17,100	27,360
Tax Credit (Deprec on Cache)	9,462	15,139	15,582	15,847
Tax on Gain on Disp of Asset	0	0	0	-19,075
TOTAL	-131,511	-2,831	-59,418	109,522
NPV (11.25%)	-92,154

Option 2: Extend lease on Disk Controller, Lease Cache Memory Upgrade

	1990 Feb-Dec	1991 Jan-Dec	1992 Jan-Dec	1993 Jan-Mar
Current Lease on Controller	-24,915	-27,180	-27,180	-9,060
Lease on Cache Memory Upgrade	-35,068	-38,256	-38,256	-12,752
SUBTOTAL	-59,983	-65,436	-65,436	-21,812
Tax Credit	22,794	24,866	24,866	8,289
TOTAL	-37,189	-40,570	-40,570	-13,523
NPV (11.25%)	-104,502

Option 3: Buyout lease on Disk Controller, Buy Cache Memory Upgrade

	1990 Feb-Dec	1991 Jan-Dec	1992 Jan-Dec	1993 Jan-Mar
Buyout Lease on controller	-101,000	0	0	0
Purchase Cache Memory Upgrade	-124,500	0	0	0
Sell Controller	0	0	0	112,750
SUBTOTAL	-225,500	0	0	112,750
Depreciation	45,100	72,160	43,296	25,978
Tax Credit (Depreciation)	17,138	27,421	16,452	9,871
Tax on Gain on Disp of Assets	0	0	0	-28,038
TOTAL	-208,362	27,421	16,452	94,584
NPV (11.25%)	-91,440

Option 4: Buy Controller and Cache Memory, Utilize XXXX Credit Corp Financing

	1990 Feb-Dec	1991 Jan-Dec	1992 Jan-Dec	1993 Jan-Mar
XXX Financing	-82,775	-90,300	-90,300	-7,525
Sell Controller	0	0	0	112,750
SUBTOTAL	-82,775	-90,300	-90,300	105,225
Tax Credit on Interest	8,621	6,247	2,535	29
Depreciation (Controller)	45,100	72,160	43,296	25,978
Tax Credit (Deprec)	17,138	27,421	16,452	9,871
Tax on Gain on Disp of Asset	0	0	0	-28,038
TOTAL	-57,016	-56,632	-71,313	87,088
NPV (11.25%)	-91,947

V. Tangible and Intangible Costs

The tangible costs of the 32 Mb cache memory upgrade to the 3990-002 controller are:

n $5,090 per month ($7,525 less $2,435 being paid currently for the 3990)

n maintenance increase from $316 to $684 per month (based on our xxxxxx agreement for maintenance which discounts maintenance on this equipment at 17%)

The intangible costs are:

n an opportunity cost if XXX releases new technology equipment to the replace the 3990 disk controller. We find this unlikely because the previous model controller, the 3880, was sold by XXX for ten years and is still available on the used equipment market.

n an opportunity cost if the response time issue ‘goes away’.

In order for the response time issue to ‘go away’, the order entry and distribution system or significant other online processing will need to ‘go away’. The portion of machine resource which the order entry and distribution system together with the general ledger system occupy, represents one quarter of the average daily transaction rate. Several systems would need to disappear in order for the order entry and distribution system to have enough resource to process at an acceptable rate. Another situation that will reduce resource usage is to remove processing from the mainframe for large business units. The processing for two business units (which have been sold) will be gone by the end of 1990. However, the principal business unit that is remaining is requesting additional processing for a new Stock Locator System, online screens for Sales Reporting, the implementation for the already defined Freight Reporting System. In addition, a newly purchased business unit’s processing will be added to the central mainframe. It appears that the difference is, at best, a wash.

VI. Tangible and Intangible Benefits.

The tangible benefits of acquiring the 32 Mb cache memory upgrade to the 3990-02 disk controller are:

n 10-20% increase in throughput on the mainframe. If throughput increases 20%, we will have the potential for processing an additional 32,000 transactions per day with the same personnel and mainframe.

n the increase in throughput translates directly to more orders entered, more goods shipped. If there are no more orders to enter then there is no benefit. But if there are more orders to be entered and the constraints of the system were preventing them from being entered, then there will be benefit. The principal business unit of the company has assured us that they will have $30,000,000 months and will need for the system to not constrain them from entering this high number of orders.

The intangible benefits are:

n the order entry and distribution system will be more likely to be utilized to its full potential.

n the data processing system will be less likely to constrain the users of any of the on-line systems

n the general ledger system and the order entry and distribution system compete head-to-head for resources. Both are resource-hungry systems. Relief is forthcoming for both systems when cache is installed. This becomes very meaning ful at month-end close. The batch allocation subsystem in the order entry system and the batch reports from general ledger cannot currently run in the eight hours of third shift on month-end close nights. As the principal business unit gets into its peak order season, the window of time form 12 pm to 6 am is not large enough for the batch processing that must occur. It is in this batch mode, where input/output bound processes compete for access to the disk devices, that the cache memory will be most helpful.

VII. Future Considerations

At this moment in time, we cannot know what our company’s systems requirements will be for the next year or five years from now. We do know that we must position ourselves so that we can run as efficiently as possible and be able to scale up or scale down as flexibly as possible. If we are asked to handle a significantly larger workload we must be able to expand our capacity quickly. The disk controller we have (3990), has the capacity to handle additional strings of disk drives. If we are asked to scale down, we are currently running close to 100% utilization on the CPU. Scaling back will not result in a glut of excess capacity.

As our utilization of the computer resource changes due to:

n continued movement from batch to on-line real-time processing,

n movement of processing from a centralized mainframe processing resource to distributed connected computers with a centralized repository for data,

n expansion/contraction of service as the company strategy unfolds,

the upgrade of the disk controller is in rhythm with these changes.

In the systems environment that we are in, we always ask ‘What is the next bottleneck?’ and ‘When will it become a problem?’. After some research, we answered , ‘The next bottleneck will be the power of the CPU’. The mainframe is a 14 mips (millions of instructions per second) engine. Freeing the disk devices to move data more efficiently will allow the operating system to drive the CPU utilization up close to 100%. When we need more throughput than the CPU can deliver at 100% utilization, we will need a bigger engine.

I leave you with a thought. Increasingly how a company does business is defined by their information systems. If a company wants to re-define how they do business, they must re-define their information systems.

APPENDIX A

Exhibit 1 Table of DASD Cache Analysis (4 p.m. composite)

Exhibit 2 Graph of DASD Cache Analysis (4 p.m. composite)

Exhibit 3 Graph of DASD Cache Analysis (10 p.m. composite)

Exhibit 4 Bar chart of DASD Cache Analysis (4 p.m. composite)

Exhibit 5 Graph of Projected Value Curve of Disk Controller

(These graphs are not included in this example)

APPENDIX B

What is Cache Memory and How Does It Help!

The disk controller is the device that receives the commands from the CPU to eigher read or write a record to or from disk. The controller is intelligent enough to know where to get the record or where to write it. One controller has the capacity to take care of several disk drives. In our installation, we only need one disk controller. The reason the cache memory is helpful is because of the differences in speed between the CPU (micro-second range, 1/1,000,000 second) versus disk devices (milli-second range, 1/1,000 second). When cache memory is available to the controller, it uses it to store data. For instance, if the CPU asks for a record, the controller gets the record for the CPU and also puts into cache memory a block of records that are in close proximity to the record asked for. Then if the CPU asks for another record, the controller searches its cache memory to see if it already has it. If it does, it passes the record to the CPU in a fraction of the time (1/1,000) required if it had to read it from disk.

When processing output without cache memory, the CPU would hand the record to be written to the controller. Then the CPU waits until the controller writes the record to the disk and returns a message to the CPU that the record has been written and it’s okay to continue processing.

With cache memory, the CPU will hand the record to the controller. The controller puts the record into its cache memory and tells the CPU that it will write it later. The CPU can go on processing without waiting for the disk access. Then the controller writes the record to disk on its own time.

Whenever the CPU is waiting on activities that operate in the microsecond range instead of millisecond range, it can process much more quickly. This is an over simplification because the CPU does multi-processing but it is the essence of why having cache memory on the disk controller makes for a considerable time saving.