Server Consolidation Examples (So Easy, a Cave Man Could Do It) Developed by: Chris Churchey Principle ATS Group, LLC & Galileo Performance Explorer 1
Methodology 1. Define hosts/servers for consolidation analysis 2. Define time-frame (Baseline) to use, related to business peaks: - Avg. 24x7 - Peak month Avg.(default) - Peak month Peaks Avg. - Avg. M-F 8am-6pm - Peak M-Tu/June-July - Single Peak of year 3. Determine CPU, Memory, Network and Disk usage for (2) above - Normalize CPU usage to IBM rperfs if Power servers, else normalize to an industry standard benchmark rating (e.g. SpecInt, SpecInt_Rate, RPE) - Consider details: Network/HBA ports, Space/Rack/CoLo, 1-core per 150K network-packets/s 4. Define Growth projection. Business provides/derive from past Trend - If unknown, I use (Baseline+25%)+15% year-to-year over 3-5 years 5. Recommend Solutions 2
Example 1 1. Define hosts/servers for consolidation analysis 129 hosts (Power5-Power7, dedicated network & hba adapters) 8 Dev (0 application / 8 Database) 50 Test (30 application / 20 Database) 71 Stage (44 application / 27 Database) Total (74 application / 55 Database) 2. Define time-frame (Baseline) Used Peak Month Average for 2014: December 1-December 31 3
Example 1: 2. Define time-frame find Peak month Created a Virtual Frame w/galileo for the 129 hosts and viewed CPU for 2014 4
Example 1: 2. Define time-frame determine Avg. Zoomed into Peak month Dec. 2014 and determine Baseline Avg. 5
Example 1 1. Define hosts/servers for consolidation analysis 129 hosts (Power5-Power7, dedicated network & hba adapters) 8 Dev (0 application / 8 Database) 50 Test (30 application / 20 Database) 71 Stage (44 application / 27 Database) Total (74 application / 55 Database) 2. Define time-frame (Baseline) Used Peak Month Average for 2014: December 1-December 31 3. Determine CPU, Memory, Network and Disk Usage 6
Example 1: 3. Determine CPU Normalized Baseline 129 hosts CPU normalized rperf Baseline Avg. usage: 690 7
Example 1: 3. Determine Memory Used 129 hosts Memory Used: 1.2TB 8
Example 1: 3. Determine Memory + Pgsp Used 129 hosts Memory Pgsp Used: 940GB Memory Baseline is Used + Pgsp Used: 1.2TB + 940GB = 2.14TB 9
Example 1: 3. Determine Network Throughput/s 129 hosts Network Throughput Baseline: 370MB/s (use Peak) 10
Example 1: 3. Determine Network Packets/s 129 hosts Network Packets Baseline: 190K/s (use Peak) These hosts are moving from Physical to Virtual adapters, so Add 2 cores to the final Solution 11
Example 1: 3. Determine Disk Throughput/s 129 hosts Disk Throughput Baseline: 1.7GB/s (use Peak) 12
Example 1: 3. Determine Disk Transfer/s 129 hosts Disk Transfer Baseline: 28,000/s (use Peak) 13
Example 1: 3. Determine Tape No Tape drives. If Tape exists, you need to account for the Tape Adapters and Throughput. Galileo collects and Trends Tape Usage for AIX hosts 14
Example 1 1. Define hosts/servers for consolidation analysis 129 hosts (Power5-Power7, dedicated network & hba adapters) 8 Dev (0 application / 8 Database) 50 Test (30 application / 20 Database) 71 Stage (44 application / 27 Database) Total (74 application / 55 Database) 2. Define time-frame (Baseline) Used Peak Month Average for 2014: December 1-December 31 3. Determine CPU, Memory, Network and Disk Usage 4. Define Growth Projection 15
Example 1: 4. Define Growth Projection Management decided (Baseline+25%)+5% year-to-year over 4 years Projected Capacity Needs based on above: CPU rperfs (690+173)+43+45+48+50 = 1,049 rperfs Memory (2.14TB+. 3.25TB Network (370MB/s+. 562MB/s Throughput Network (190K/s+ 288K/s Packets Disk (1.7GB/s+.. 2.6GB/s Throughput Disk (28K/s+. 42.5K/s IOps Tape none 16
Example 1: 5. Recommend Solution CPU rperfs 1,049 rperfs - 3-4 S824 s (each 421 rperfs) depending upon if all 129 Hosts need to be available if a Server fails - 2 E870 s (856-1350 rperfs) Memory 3.25TB - 1.5-2TB on each S824-2-3.5TB on each E870 depending upon if one needs to support all 129 Hosts if a Server fails Network 562MB/s Throughput - Prefer 10Gb adapters, determined by # of VIO s and if redundancy within VIO is desired *Optionally several 1Gb adapters/ports also determined on the above *10Gb = 850-900MB/s per Port vs. 1Gb = 80-100MB/s per Port Network 288K/s Packets Disk 2.6GB/s Throughput - 8-16Gb adapters, determined by # of VIO s and if redundancy within VIO is desired *8Gb = 800-850MB/s per Port Disk 42.5K/s IOps 17
Example 2: 272 Power5-7 Servers (530+ LPARs) Mostly dedicated servers - Totaled installed 20,879 rperfs capacity, 9TB+ memory, 1,368 FC ports and 3,194 Network ports - They were told to buy 50+ S824 s to have equiv. capacity * Installed doesn t mean it is used or necessary! * Why you do a Server Consolidation assessment easily with Galileo PE After analysis, looking at Time-overlap usage of CPU, Memory, Network, Disk and Tape: - They needed 28 S824 s * Will implement several LPM groupings based on separation requirements * Will 100% virtualize using VIO, SEA, NPIV, MSPP and reduce the Port/Cable counts to 10-15% of what it was They are evaluating E870 for further reduction 18
Questions? We can help analyze and implement. Contact us! Online: www.galileosuite.com Email: sales@galileosuite.com Phone: +1-484-320-4302 19