Systems Integration Outsourcing Infrastructure Server Technology Consulting unisys

  • Slides: 24
Download presentation
Systems Integration. Outsourcing. Infrastructure. Server Technology. Consulting. unisys Imagine it. Done. The Thoroughly Modern

Systems Integration. Outsourcing. Infrastructure. Server Technology. Consulting. unisys Imagine it. Done. The Thoroughly Modern Mainframe Dr. Michael Salsburg NTSMF Users' Group Dec 9, 2002

Agenda § Large Scale WINTEL Servers § Disruptive technology or trend? § Scale Up

Agenda § Large Scale WINTEL Servers § Disruptive technology or trend? § Scale Up or Scale Out ? § A Workload-motivated discussion of SMP and CC-NUMA § PCI-Based I/O § Consolidation § Emerging Technologies unisys Imagine it. Done.

Server Industry Trends Source: IDC Intel will dominate server chip market Windows 2000 will

Server Industry Trends Source: IDC Intel will dominate server chip market Windows 2000 will be pervasive server OS unisys Imagine it. Done.

The x 440 Competition Gartner Oct 2002 Unisys ES 7000 Aries 230 Processors Supported

The x 440 Competition Gartner Oct 2002 Unisys ES 7000 Aries 230 Processors Supported Unisys ES 7000 Orion Egenera Blade Frame HP Pro. Liant DL 760 Intel Xeon 1. 4, 1. 6 Pentium 1. 4, 1. 6 GHz 700, 900 GHz MHz HP rp 8400 IBM e. Server x 440 HP PA- Intel Xeon 8700 1. 4, 1. 5, 750, 875 1. 6 MHz GHz Max Procs 16 32 96 8 16 8 , 16 by 2002 Max Mem 32 G 64 G 288 G 16 G 64 G Max PCI Slots 48 96 12 10 16 72 unisys Imagine it. Done.

A Comparison using Moore’s Law § Comparison of CPU Speeds / tpc. M for

A Comparison using Moore’s Law § Comparison of CPU Speeds / tpc. M for 4 x cpu WINTEL systems unisys Imagine it. Done.

TPC-C Top 10 unisys Imagine it. Done.

TPC-C Top 10 unisys Imagine it. Done.

unisys Imagine it. Done.

unisys Imagine it. Done.

Scale Up or Scale Out? § Two of the 3 -tiers in current application

Scale Up or Scale Out? § Two of the 3 -tiers in current application architectures use scale-out for growth § Increase # of Web servers § Increase # of Application Servers § Database back end cannot be scaled out § Scale up is needed for large database applications § Scale out has some inherent down sides § additional administrative/management attention § Move “headroom” needed for heavy traffic unisys Imagine it. Done.

SMP / NUMA Workload Discussion § As code executes on the processor, memory is

SMP / NUMA Workload Discussion § As code executes on the processor, memory is referenced. This can be broken into three regions § High Locality of Reference § Memory is immediately re-referenced (> 95%) § Working Set – the set of addresses on which the software primarily focuses § Persistent Storage – addresses that are stored on physical devices unisys Imagine it. Done.

Scale Out- SMP or NUMA? Workload Interference § When two processes are running on

Scale Out- SMP or NUMA? Workload Interference § When two processes are running on the same system, their memory references will interfere. § It is preferable to only interfere at the persistent storage level § Interference at higher levels can decrease cache efficiency and slow down processing, effectively reducing the CPU power unisys Imagine it. Done.

SMP / NUMA SMP Topology § A bank of CPUs share a bank of

SMP / NUMA SMP Topology § A bank of CPUs share a bank of Memory § Each CPU has a local cache to optimize high locality of reference § A cache miss has uniform latency time to get data from memory § “Dirty” memory references require fetching the updated memory from another CPU’s cache § The CPU can “stall” waiting for a memory reference unisys Imagine it. Done.

SMP / NUMA Workload Discussion § Percentages of references based on TPC-C workload profile

SMP / NUMA Workload Discussion § Percentages of references based on TPC-C workload profile § Relative time units show orders of magnitude between cache hit and persistent storage unisys Reference Level Time Units Reference Level Percentage CPU Cache 1 CPU Cache 98. 0% Main Memory 100 Main Memory 01. 9% Remote Cache 200 Persistent Storage 00. 1% Persistent Storage 10, 000 Imagine it. Done.

SMP / NUMA (Non-Uniform Memory Access) § Overcome bus congestion and physical fabrication limitations

SMP / NUMA (Non-Uniform Memory Access) § Overcome bus congestion and physical fabrication limitations found in a single bus architecture § Two memory latencies – near and far § The NUMA ratio is the ratio of far latency over near latency § Originally 30, now it is around 3 unisys Imagine it. Done.

SMP / NUMA Hybrid (Unisys ES 7000) § Another level of cache is introduced

SMP / NUMA Hybrid (Unisys ES 7000) § Another level of cache is introduced § Memory accesses can be non-uniform when comparing Next Level Cache hits to memory references § Overcomes the fabrication/congestion problems of a single bus architecture unisys Imagine it. Done.

PCI-Based I/O Cellular Multi. Processing (CMP) Architecture unisys Imagine it. Done.

PCI-Based I/O Cellular Multi. Processing (CMP) Architecture unisys Imagine it. Done.

6. 4 GB SP 2 PCI-Based I/O ilit b ala c S sport ess

6. 4 GB SP 2 PCI-Based I/O ilit b ala c S sport ess r. Tran Hype GB/sec Max Bus or per direction t 533 MHz 4 3 I-X PC r xp E I- PC 16 X, or 8 X@5 Gb SP 1 266 2 1 X 4 X 8 X 133 1 66 unisys or y. P 0. 8 GB I PC 2001 & earlier 2002 Imagine it. Done. 2003 2004 2005

Enterprise-Level Backup / Restore unisys Imagine it. Done.

Enterprise-Level Backup / Restore unisys Imagine it. Done.

Enterprise-Level Backup / Restore § Complete recovery of a 2. 5 terabyte database: §

Enterprise-Level Backup / Restore § Complete recovery of a 2. 5 terabyte database: § From tape, the database was recovered in only 88 minutes with a sustained throughput during restore of 2. 2 TB/hr. § From the hardware snapshot, the same database was recovered in only 11 minutes. § Complete backup of a 2. 5 terabyte database: § Backup to tape took only 68 minutes with minimal impact on online operations and sustained throughput of 2. 6 TB/hr. unisys Imagine it. Done.

Consolidation § "[Our] servers were multiplying like rabbits, " says Jeff Smith, manager of

Consolidation § "[Our] servers were multiplying like rabbits, " says Jeff Smith, manager of corporate network services at La-Z-Boy Inc. , a Monroe, Mich. -based residential furniture producer that just completed a Windows NT server consolidation project. "Our distributed environment was becoming more and more difficult to manage. " § Thinning The Server Ranks Computerworld Aug 26, 2002 unisys Imagine it. Done.

Consolidation § How do you stuff over 130 CPUs’ worth of workload into a

Consolidation § How do you stuff over 130 CPUs’ worth of workload into a 32 x CPU system? § Veeerrrry carefully…… § Why are current server farms filled with under-utilized servers? § Web Hosting Sites § “New web servers are installed when Peak CPU utilization reaches above 35%. ” § “Speed and reliability are very important to your web site. All of our servers are maintained at less than 15% CPU utilization. This ensures that your web site downloads as fast as possible!” unisys Imagine it. Done.

Consolidation Responsive Consolidation § Which would you prefer – an average queue size of

Consolidation Responsive Consolidation § Which would you prefer – an average queue size of 0. 2 on a 1 x or a 32 x system? unisys Imagine it. Done.

Consolidation Benefits § § § Simplified Management / Administration Higher Utilization (less “headroom”) Less

Consolidation Benefits § § § Simplified Management / Administration Higher Utilization (less “headroom”) Less Variability of Service Less Overall CPU Overhead Less software licenses unisys Imagine it. Done.

Emerging CPU Technologies 32 x INTEL CPU TPC-C Results Date Published tpm-C Chip Speed

Emerging CPU Technologies 32 x INTEL CPU TPC-C Results Date Published tpm-C Chip Speed Cache Memory 11/11/2001 165, 218 Pentium III 900 MHz Xeon 2 MB 64 GB 9/9/2002 308, 620 Itanium II 1 GHz 3 MB 256 GB 11/4/2002 203, 518 Pentium IV Xeon 1. 6 GHz 1 MB 64 GB unisys Imagine it. Done.

Itanium II What’s so great about 64 bits? § For transaction processing, memory addressing

Itanium II What’s so great about 64 bits? § For transaction processing, memory addressing is increased and therefore the amount of main memory increases § The top 5 TPC-C results were achieved using 64 bit computing § TPC-C is a large database application – this is a sweet spot for 64 bit commercial computing Bigger is DEFINITELY Better!! unisys Imagine it. Done.