Hardware evaluation and procurement Hardware: competition, - - PowerPoint PPT Presentation

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Hardware evaluation and procurement

Evaluation of CPU nodes for procurement purposes at CC-IN2P3 Yannick Perret

Centre de Calcul IN2P3

3T1 22-23 May 2007

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Outline

1

Hardware: competition, evolution, manufacturers. . . Hardware: CPU Manufacturers

2

Evaluation Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

3

Hardwares procurements

4

Some results

5

Conclusion

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Outline

1

Hardware: competition, evolution, manufacturers. . . Hardware: CPU Manufacturers

2

Evaluation Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

3

Hardwares procurements

4

Some results

5

Conclusion

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

CPU evolution

An evolving situation ◮ AMD and Intel produce good *x86 processors ◮ CPU power increases very fast

◮ number of cores ◮ frequencies

◮ Electric needs are decreasing ◮ new CPU frequencies (for a given model) ◮ new CPU models each 6 (or few) mounths Situation is not stable, competition exists, models evolve.

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Manufacturers / retailers

Known companies: DELL, HP, IBM, NEC, SUN, . . . Roughtly the same machines ◮ same processor(s) ◮ same motherboard architecture ◮ same memory With some differences ◮ disks technology, integrated RAID ◮ management (IPMI, Service processor, . . . ) ◮ quality (air cooling, power supply, . . . ) ◮ support quality ◮ integration (rack and/or box) ◮ . . . and the price. . .

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Manufacturers / retailers

Known companies: DELL, HP, IBM, NEC, SUN, . . . Roughtly the same machines ◮ same processor(s) ◮ same motherboard architecture ◮ same memory With some differences ◮ disks technology, integrated RAID ◮ management (IPMI, Service processor, . . . ) ◮ quality (air cooling, power supply, . . . ) ◮ support quality ◮ integration (rack and/or box) ◮ . . . and the price. . .

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Outline

1

Hardware: competition, evolution, manufacturers. . . Hardware: CPU Manufacturers

2

Evaluation Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

3

Hardwares procurements

4

Some results

5

Conclusion

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Experiments

Experiments have various needs ◮ pure CPU jobs ◮ large I/O jobs (disk and/or network) ◮ large memory usage ◮ a combination of items We must take all these needs in account

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

CC-IN2P3 needs

CC-IN2P3 has its own needs ◮ to reach or computing goals ◮ minimize needed electric power ◮ minimize heat generation ◮ minimize used floor space in the computer room ◮ minimize cost!

◮ electric supplies cost ◮ network supplies cost ◮ electric consumption cost

• So. . . to obtain the best at the lowest cost!

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Configuration constraints

Moreover, some choices imply constraints The number of cores implies configuration constraints ◮ needed amount of memory ◮ disk size ◮ network bandwidth

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Existing benchmarks

CPU and machines manufacturers give benchmarks of CPU power and electric consumption (SPECint and electric consumption) This is not usable ◮ CPU manufacturer benchmarks are more and more irrelevant ◮ real jobs rarely optimize ressources of a given CPU ◮ maximum electric consumption is not the effective consumption

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Our benchmarks

Our benchmarks are based on an estimation of real-world jobs ◮ CPU medium to intensive treatments ◮ medium to large memory usage ◮ disk and network I/Os In addition electric consumption is measured in various situations

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Our benchmarks (2)

Measured values ◮ CPU power is computed as the power ratio of a reference machine ◮ power consumption is also computed as the power ratio ◮ effective minimal, maximal and average power consumption ◮ CPU effectiveness is the ratio between the previous values (amount of electric power needed to compute the same amount of treatments) We urge manufacturers to lend us new machines. . .

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Outline

1

Hardware: competition, evolution, manufacturers. . . Hardware: CPU Manufacturers

2

Evaluation Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

3

Hardwares procurements

4

Some results

5

Conclusion

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Hardwares procurements

For hardware procurement, we give the ratio between measured and theoretical SPECint values for CPU models. Offers must match the total CPU power needed (with our measures), the maintenance requirements and the other constraints (e.g. management) Then additionnal costs are added to the offer ◮ cost of the electric consumption for 3 years ◮ electric supplies (depend on number of machines and consumption) ◮ network supplies (depend on number of racks) Some other points, as the total required floor space, are not directly evaluated.

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Outline

1

Hardware: competition, evolution, manufacturers. . . Hardware: CPU Manufacturers

2

Evaluation Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

3

Hardwares procurements

4

Some results

5

Conclusion

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Some results

Benchmarks on some recently tested machines

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Some rules

Some rules from the tests ◮ CPU power is linear to number of cores ◮ more cores means better efficiency ◮ efficiency is the same for all frequencies ◮ CPU power depends only on processors ◮ integration and quality can induce power consumption variations

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Outline

1

Hardware: competition, evolution, manufacturers. . . Hardware: CPU Manufacturers

2

Evaluation Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

3

Hardwares procurements

4

Some results

5

Conclusion

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

Conclusion

Evaluation is important for procurements ◮ before procurements (to know what we can get and define specifications) ◮ after procurements (to validate offers) Evaluation is important for computer rooms ◮ planification of electric needs ◮ planification of coolling systems ◮ planification of floor space usage An everyday job Evaluating CPU nodes is a recurrent job.

Yannick Perret Hardware evaluation and procurement

Hardware evaluation and procurement Yannick Perret Hardware: competition, evolution, manufactur-

• ers. . .

Hardware: CPU Manufacturers

Evaluation

Whom for, what for? CC-IN2P3 Existing benchmarks Our benchmarks

Hardwares procurements Some results Conclusion

A part of. . .

Yannick Perret Hardware evaluation and procurement