ACROFAN

Dell PowerEdge R730 Rack optimized server : Performance

Published : Tuesday, September 16, 2014, 9:53 am
ACROFAN=Yong-Man Kwon | yongman.kwon@acrofan.com | SNS
2U, 2socket x86 server used by almost all application workload, so need to prepare not only performance and efficiency by increase service request but flexible for matching various workloads.

Many modern 2U servers focused to performance and scalability not only of processor and memory but storage subsystems, suggest many ways to make system for optimization of various workloads. Dell PowerEdge R730 has strong characteristic of that kind of flexibility.

 
▲ Test system configurations

Test system has best class of performance expect in 2U, 2socket server. Intel Xeon E5-2690 v3 provide 12cores, 24threads and 2.6GHz operation speed per processor, 24cores 48threads per overall system. Memory subsystems configured to 64GB DDR4 ECC quad channels, and storage configured to RAID 6 by 5 disk of 10,000rpm 2.5inch SAS disk.

Red Hat Enterprise Linux 6.5 was used to test system’s operating system and installed by OS deployment function in dell lifecycle controller because it’s one of most stable OS environment and certified by Dell. Test configured to measure arithmetic and storage performance, and service performance depends on basic arithmetic and storage performance. The results can compare to other results of previous generations tested by same tools.

■ High performance for more various workload situations.

 
▲ Test results of Intel Linpack 11.2, Unit MFlops, Higher is better

 
▲ Test results of OpenSSL 1.0.1g, Unit Signs/sec, Higher is better

Linpack test results that one of important performance guideline in HPC, is significantly improve performance than previous generations. Results of intel MKL 11.2 that use processor’s AVX2 instructions efficiently, can compared to performance of 4socket server used to previous processors. This results shows productivity improvements of new server with new processor and platform than previous if you want to high arithmetic performance and software supports to new instructions such as AVX2.

One interesting trend in results of linpack test is need huge workloads for peak performance in this test system. This test system shows near the peak performance in problem size 30,000 or over, and it is higher than previous generation that usually shows peak performance near problem size 20,000. This result also means that require huge and parallel workloads for utilize this system most efficiently and sometimes don’t have performance advantages depends on the workload conditions.

Results of OpenSSL encryption performance based on RSA 4096bit shows high performance levels. Test system’s Intel Xeon E5 processor supports AES-NI and hardware encryption acceleration by processor, and support high performance in encryption with low processor load if utilize this feature. And this system’s high encryption performance affects many ways to company’s IT environments not only users who use encryption but OEM companies for make security appliance based on dell’s hardware.

 
▲ Test results of RAMSpeed SMP 3.5.0, Unit MB/s, Higher is better

 
▲ Test results of DBench, Unit MB/s, Higher is better

Because of having maximum 18 cores per processor in latest 2socket platform, analyzing system performance is not simple any more. Systems do not run their full performance when the workload is heavy enough to such as the test results of intel linpack. When we try to measure memory performance, STREAM that a traditional test tools can’t measure latest system’s memory performance correctly. In results of STREAM or other traditional test tools have huge variations or incorrectly.

Result of memory performance using RAMSpeed SMP, it is not enough we expect only consider the values but it’s right results compare with other system’s results. The results are similar to previous DDR3 based memory subsystems, but performance differentiation between DDR3-1600 and DDR4-2133 is maximum 20% theoretically. DDR4 is only early stage in use and have more potential in performance when speed up to DDR4-3200 or more we expected.

Result of Dbench test for measuring file service performance on the storage can check overall performance of storage system with disk and RAID controller. Peak performance in test system is 4648MB/s at 48 users on test storage configured to RAID 6 array by 5ea of 10k rpm SAS drives and PERC H730P raid controller with 2GB cache. Of course due to cache system on controller and OS, but still over 3GB/s of transfer performance at 256 users is very impressive performance and user can feeling good performance in real usage.

 
▲ Computing Benchmark Series, Higher is better in results.

 
▲ Timed Benchmark Series, Unit sec, Shorter is better in results

Overall results on various series of tests shows improve both of performance significantly such as peak performance and usual performance on low workloads. In result of FFTE or Himeno benchmarks, R730 has improved the peak performance and performance efficiency depends on operating speeds. This is important advantages by new processor and platform in new system.

R730 get significant improvements in test results of database or sort than results of previous generations. Results of HMMer search or MAFFT alignment tests depends on processor core’s efficiency is higher 2 times or more than previous Xeon 5600 series, even quite higher than previous Xeon E5-2600 series based systems.

Results of pgbench tests based on PostgreSQL, R730 has improved performance than previous because results affected by the balance of test system configuration. Results of PostMark for testing disk transaction performance is not much improved than pgbench, so disk performance is not a reason of performance improvement about SQL. Better balance with improve performance of processor and memory make better SQL performance, and it can see that test system was complete to run the MySQLBench spend only 350sec.

■ New generation 2U-level server that try to do everything

Recently X86 processor based server have enough performance to use for various services not only just cost performance, and applying more widely from simple web service to server cluster, HPC, mission critical system in companies that is more complex and can’t use in times past. And 2U form factor rack server is using most popular because have good balances of cost performance and extensibilities.

Flexibility for maximize utilization in 2U form factor is one of special features of Dell PowerEdge R730. Outstanding system design for maximum flexibility with high performance and efficiency from new processor and platform makes this system more powerful. For example, R730 support to using high performance GPU for GPGPU that have limitation in use previous systems because space, cooling and power supply. Also in storage, Dell prepare to support high capacity storage configuration with flash storage or suggest special model that is specialized with storage.

Flexibility in platform and system of Dell PowerEdge R730 can optimize to any workloads and deliver maximum performance and efficiency to users. In modern datacenter, configuration of hardware infrastructure required to more ‘simple’ for managements even software environment become more complicated. In this situation, flexibility of Dell PowerEdge R730 that can configure all kind of optimize system become the outstanding point of R730 with well-make management tools in whole infrastructure.

 
▲ Product specification of Dell PowerEdge R730 rack optimized server


Copyright © acrofan All Right Reserved


    Acrofan     |     Contact Us : guide@acrofan.com     |     Contents API : RSS

Copyright © Acrofan All Right Reserved