Unconfigured Ad Widget

Collapse

Announcement

Collapse
No announcement yet.

Getting Started with Performance Analysis

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Getting Started with Performance Analysis

    ** This thread discusses the article: Getting Started with Performance Analysis **
    ** This thread discusses the Content article: Getting Started with Performance Analysis **
    0

  • #2
    Getting Started with Performance Analysis

    ** This thread discusses the article: Getting Started with Performance Analysis **
    I upgraded from 42 disk drives down to just 7 disk arms. These newer disk drives, and their cache cards, gave me much better performance. What worries me about recommending too many disk drives initially is the following may happen: Customer may order lots of a particular model to meet an outdated disk arm theory. Then when they do need to upgrade for space reasons, they have no space available in their new box for more drives. Now, if they updated as needed, they could take advantage of newer disk drives with better speeds, cache, etc. More details: i5/OS V5R3M0 and the new 520 The effects of upgrading an 820-24AA to a 520-7453 at Group Dekko Services, LLC “From” machine Resource Type-model Text CEC01 9404-820 Main Card Enclosure PN01 247F System Control panel MP01 25BC-0000 System Processor Card PV02 2436-001 Processor Capacity Card PV01 1523-001 Interactive Card … MS01 3002-000 128MB Main Storage Card MS02 3002-000 128MB Main Storage Card MS03 3002-000 128MB Main Storage Card MS04 3002-000 128MB Main Storage Card MS05 3005-000 512MB Main Storage Card MS06 3005-000 512MB Main Storage Card MS07 3004-000 256MB Main Storage Card MS08 3004-000 256MB Main Storage Card … Tape card = 6534 Tape Drive = 3590E Disk 6714’s 4 6718’s 4 6607’s 30 6719’s 4 Total 42 Single partition of i5/OS. System Console: Twinax “To” machine Resource Type-model Text CEC01 9406-520 Main Card Enclosure PN01 28E5 System Control Panel MP01 522A System Processor Card PV01 7453 Processor Capacity Card SP01 28D7 Service Processor Card … MS01 30D3 1024MB Main Storage Card MS02 30D3 1024MB Main Storage Card MS03 30D3 1024MB Main Storage Card MS04 30D3 1024MB Main Storage Card Tape card = 2749 Tape drive = 3590E Disk 4327’s 7 Total 7 Two partitions of i5/OS. A vast bulk of the equipment is in the first partition. Nothing has been done with the second partition except loading i5/OS. System Console: Hardware Management Console, or HMC. Application Base Roughly 11 developers primarily developing primarily traditional RPG applications using a mix of 5250 and C/S based development tools. Three Domino servers used for testing and development. Some minor training of applications. Conversion process An unload/reload from the old machine to the new machine was completed using the “Backup and Recovery Guide” – Appendix D “Recovering your server to a different server”. http://publib.boulder.ibm.com/infoce...u/sc415304.pdf Time trial #1: Time to do a complete system save Both systems used the same 3590E tape drive. Broke down GO SAVE 21 by it’s components. To save all 2009 libraries on the 820 took 6 hours and 29 minutes. To save all 2009 libraries on the 520 took 3 hours. To save all 17,875 DLO’s on the 820 took 5 minutes. To save all 17,875 DLO’s on the 520 took 3 minutes. To save 161,969 IFS objects on the 820 took 57 minutes. To save 162,033 IFS objects on the 520 took 29 minutes. Don’t know if it: 1) was the difference between the 6534 and the 2749 card 2) the way we have them configured (do we have the tape card on the 520 not sharing a processor with disk and not the same on the 820?) 3) or the difference between the 520 and the 820 Time trial #2: Time to do a reclaim storage, or RCLSTG Traditionally the RCLSTG on the 820 took between 13 and 14 hours. And we run it every 8 weeks. The RCLSTG on the 520 took 6 hours and 30 minutes. Time trial #3: Time to do nightly save Doing the nightly backup on the 820 using the 6534 card and the 3590E varied between 8.5 to 9 hours. Doing the nightly backup on the 520 using the 2749 card and the 3590E only took 3.5 hours. Time trial #4: Applications Actually our compiles and other applications weren’t running bad. When doing training, etc., we were occasionally hitting the wall in interactive processing. But just for you, we tried a SSA-BPCS period end close. Just to refresh your memory, we did the unload/reload and ran the period end close on both machines, with identical data. On the 820 it took POC512 13 seconds INV900 3 hours, 8 minutes and 57 seconds Reorg ITH 24 minutes and 43 seconds Total 3 hours, 33 minutes and 53 seconds On the 520 it took POC512 6 seconds INV900 50 minutes and 16 seconds Reorg ITH 12 minutes and 29 seconds Total 50 minutes and 21 seconds (PS: The 820 was committed to a single user machine because all external cables to the lan, etc., were moved over to the 520. Leaving only the twinax console. The 820 was NOT in a restricted state however. The three Domino servers and other tasks were running. The 520 was running in normal mode in the middle of the day with programmers banging out code, and me using the Domino for my email, etc.) Benefits of upgrade: · Dropped from a P30 to a P10 resulting in lower software maintenance costs. · Faster speeds · Less downtime due to backup · Lower hardware maintenance costs · Smaller footprint. Replaced a system unit, expansion unit, twinax terminal, and rack for the 3590 with a single rack that contained all of the following: system unit, expansion unit, HMC, 3582-LTO, and more… Later upgrades performed and their results Replaced 2749 card and 3590E tape drive with a 5704 fiber card attached to a 3582 LTO2 tape drive via a 2109-F16 fiber tape switch. Results: Nightly save on the new 520 dropped from 3 hours 27 minutes down to 3 hours and 7 minutes. (On other processor combinations our tests have shown that the switch from the 3590E to the 3582 LTO-2 had a much greater impact. Actual comparison data available upon request. See also: http://archive.midrange.com/domino40.../msg00001.html http://archive.midrange.com/midrange.../msg01516.html ) Comments from the team here: I noticed the HTTP server administration screen worked a lot faster! (T.M.) Compiles, Softlanding’s TurnOver promotions, and WDSC server side tasks (such as opening a member) all run faster. (D.S.) WRKOBJ was much faster, especially noticeable when run over *ALL libraries, *ALL object types, and for a generic object name. (D.M.)

    Comment


    • #3
      Getting Started with Performance Analysis

      ** This thread discusses the article: Getting Started with Performance Analysis **
      robberendt wrote: > I upgraded from 42 disk drives down to just 7 disk arms. These newer > disk drives, and their cache cards, gave me much better performance. Let's be a little realistic here, 1 hard drive will perform better than a million floppy drives, and that's almost the situation you are dealing with. You're comparing an I5 machine with 15k drives against an older I/O setup and much slower drives. I would agree with drive arms being a myth if you compared 14 of the 35GB's against the 7 70GB's and didn't see a difference. Bill

      Comment

      Working...
      X