<<< DOCD$:[NOTES$LIBRARY]SCT-RAVEN.NOTE;1 >>> -< The SCT Raven conference >- ================================================================================ Note 794.0 RAVEN_FT1: SCS Directory lookup to HSJs by conma No replies EVMS::PERCIVAL "{VDE SCT}" 542 lines 20-JUN-1997 09:13 -------------------------------------------------------------------------------- Title: SCS Process Polling, Connection Manager polling HSxxs Project: Bugfix Author: Ian Percival Component being changed: VAX SYSLOA FACILITY CNXMAN.MAR AXP SYSLOA FACILITY CNXMAN.MAR Images affected: AXP: SYS$CLUSTER.EXE VAX: SYS$CLUSTER.EXE Date: 09-June-1997 1. Change Information: Type of change: Bug fix.. Where is the change being made: VAX/AXP RAVEN_FT1 Reviewed by: John Croll Dennis Van Eron Kevin Jenkins Greg Jordan Phil Norwich Checkin information: 2. Symptom Description: SCS will poll all disk controllers such as HSJ50s, etc, looking for the VMS$VAXCLUSTER process name. As these controllers will never have the connection manager code loaded, the lookups should be disabled. This is a small overhead as each node will only poll every PRCPOLINTERVAL which is defaulted to 30 seconds. Nevertheless, in a large cluster with many controllers and many nodes, there will be a certain amount of unnecessary interconnect traffic and high IPL processing. 3. Problem Statement: In module CNXMAN.MAR the initialisation routine, CNX$INIT, performs a universal poll enable by calling SCS$POLL_MODE with R2=0. This tells SCS to search every system block it knows about for a listening connection manager sysap. 4. Goals of this change: QAR EVMS_C_DESIGN #169 will be closed after this change. Disk controllers will no longer be polled for a sysap which will never be loaded on them. 5. Known Restrictions / Possible Limitations / Risks: Controllers which do NOT load the MSCP server will continue to be polled in the same way as they are today. This has no effect on VAX or AXP nodes which do not load the MSCP SERVER - it is only a restriction for disk controllers. 6. Design / Algorithms: Use SCS to poll for MSCP$DISK on remote nodes. Whenever it is located, determine whether the remote node is running VMS - or whether it is not. If it IS running OpenVMS, then turn of polling for the MSCP$DISK - the normal connection formation mechanisms for VMS$VAXcluster will still apply. If it is NOT running OpenVMS, then the conenction manager should never poll it, so turn off polling for both MSCP$DISK and VMS$VAXCLUSTER. 7. Testing: Using the new driver, use SDA to check the SB$B_ENBMSK field along with the appropriate polling bit in the SPPB located on the SCS$GQ_POLL queue. All controllers should have the VMS$VAXcluster polling bit cleared, once a state transition has occurred for a VAX or AXP in the cluster. 8. Documentation: None 9. Implementation: For AXP:- ************ File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;22 2 .IDENT 'X-19' 3 ; ****** File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 2 .IDENT 'X-18' 3 ; ************ ************ File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;22 40 ; X-19 JIP0011 Ian Percival 09-Jun-1997 41 ; Prevent unnecessary process polling overhead by removing 42 ; polls to disk controllers. 43 ; 1. Add new section to CNX$INIT to poll for MSCP$DISK 44 ; 2. Add new routine CNX$NEW_DISK_SYSTEM 45 ; 3. Add new routine CNX$STOP_DISK_POLL and calls to it. 46 ; 47 ; X-18 DAG Doug Gordon 22-Aug-1996 ****** File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 40 ; 41 ; X-18 DAG Doug Gordon 22-Aug-1996 ************ ************ File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;22 466 $SPPBDEF ; SCS Process Polling Block 467 $SSDEF ; Status codes ****** File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 460 $SSDEF ; Status codes ************ ************ File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;22 498 MSCP_PROC_NAME: ; SYSAP name for MSCP server 499 .ASCII "MSCP$DISK" 500 .REPT 16-<.-MSCP_PROC_NAME> 501 .ASCII " " 502 .ENDR 503 504 CLU$GL_ACKLIM:: .LONG 0 ; Calculated ACKLIM ****** File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 491 CLU$GL_ACKLIM:: .LONG 0 ; Calculated ACKLIM ************ ************ File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;22 754 755 MOVAB L^CNX$NEW_DISK_SYSTEM,R0; Address of new disk system routine 756 ; R1 still contains the CONMAN SPPB 757 ; Store this in the new SPPB context 758 ; field for later use... 759 MOVAB L^MSCP_PROC_NAME,R2 ; Address of process name 760 JSB G^SCS$POLL_PROC ; Poll for copies of MSCP Server 761 CLRL R2 ; All systems at this time... 762 JSB G^SCS$POLL_MODE ; Enable polling 763 ; Do not worry about failure of the 764 ; disk polling - this will only leave 765 ; things the same as the Day 1 766 ; pre-Raven situation. 767 768 BSBW CNX$CON_INIT ; Initialize configuration manager ****** File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 741 BSBW CNX$CON_INIT ; Initialize configuration manager ************ ************ File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;22 990 BSBW CNX$STOP_DISK_POLL ; Disbale polling this system for disks 991 POPR #^M ; Restore connection message addr ****** File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 963 POPR #^M ; Restore connection message addr ************ ************ File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;22 1448 BSBW CNX$STOP_DISK_POLL ; Disbale polling this system for disks 1449 MOVB #CSB$K_OPEN,- ; Mark connection open ****** File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 1420 MOVB #CSB$K_OPEN,- ; Mark connection open ************ ************ File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;22 2823 .SBTTL CNX$NEW_DISK_SYSTEM - Detect new disk controllers 2824 ;++ 2825 ; FUNCTIONAL DESCRIPTION: 2826 ; 2827 ; CNX$NEW_DISK_SYSTEM - Detects whether the node in question is 2828 ; a VMS node or is a disk controller. If the new system is solely 2829 ; a disk controller, polling is disabled for both the connection 2830 ; manager and the MSCP$DISK service. This reduces the anount of 2831 ; polling overhead the connection manager imposes. 2832 ; 2833 ; CALLING SEQUENCE: 2834 ; 2835 ; JSB CNX$NEW_DISK_SYSTEM - we are called here by the SCS Directory 2836 ; lookup services. 2837 ; 2838 ; INPUT PARAMETERS: 2839 ; 2840 ; R0 is context longword - we use it for storage of the CONMAN SPPB 2841 ; R1 is address of process name 2842 ; R2 is address of system ID 2843 ; R3 is address of process information 2844 ; R4 is address of node name 2845 ; 2846 ; 2847 ; IMPLICIT INPUTS: 2848 ; 2849 ; NONE 2850 ; 2851 ; OUTPUT PARAMETERS: 2852 ; NONE 2853 ; 2854 ; IMPLICIT OUTPUTS: 2855 ; SCS Polling will be disabled to non-OpenVMS nodes. 2856 ; 2857 ; COMPLETION CODES: 2858 ; 2859 ; R0 - 1 = disable polling to this system 2860 ; 0 = continue polling this system 2861 ; 2862 ; SIDE EFFECTS: 2863 ; 2864 ; 2865 ;-- 2866 CNX$NEW_DISK_SYSTEM: ; 2867 .JSB_ENTRY INPUT=,OUTPUT=,SCRATCH= 2868 MOVL R0,R4 ; save the SPPB for CONMAN 2869 MOVL R2,R3 ; Save the System ID address 2870 MOVL R2,R1 ; Address of System ID 2871 CLRL R2 ; No output buffer 2872 JSB G^SCS$CONFIG_SYS ; Locate the System Block for this node 2873 BLBC R0,20$ ; R1 = SB on success 2874 CMPL SB$T_SWTYPE(R1),- ; Is this a VMS node? If so disbale 2875 #^A"VMS " ; polling ourselves.... 2876 BNEQ 10$ 2877 MOVZBL #SS$_NORMAL, R0 ; Disable polling to this node 2878 RSB ; CONMAN will form its own connection 2879 2880 10$: CLRL R0 ; Disable Polling 2881 MOVL R4,R1 ; SPPB 2882 MOVL R3,R2 ; System ID address 2883 JSB G^SCS$POLL_MODE ; Stop CONMAN polling a Disk Controller 2884 RSB ; Success disables poll for MCSP$DISK 2885 2886 20$: CLRL R0 ; Continue to poll on failure 2887 RSB 2888 2889 2890 .SBTTL CNX$STOP_DISK_POLL - Stop looking for Disk Controllers 2891 ;++ 2892 ; FUNCTIONAL DESCRIPTION: 2893 ; 2894 ; CNX$STOP_DISK_POLL - This routine is called when a connection 2895 ; has been formed between this node and another's connection 2896 ; manager. This means that we have established the identity of the 2897 ; remote node - and so no longer need to poll for MSCP$DISK. 2898 ; 2899 ; CALLING SEQUENCE: 2900 ; 2901 ; JSB CNX$STOP_DISK_POLL 2902 ; 2903 ; INPUT PARAMETERS: 2904 ; 2905 ; R2 is address of system ID 2906 ; 2907 ; 2908 ; IMPLICIT INPUTS: 2909 ; NONE 2910 ; 2911 ; OUTPUT PARAMETERS: 2912 ; NONE 2913 ; 2914 ; IMPLICIT OUTPUTS: 2915 ; SCS Polling for MSCP$DISK will be disabled to remote nodes. 2916 ; 2917 ; COMPLETION CODES: 2918 ; NONE 2919 ; 2920 ; SIDE EFFECTS: 2921 ; 2922 ; 2923 ;-- 2924 CNX$STOP_DISK_POLL: ; 2925 .JSB_ENTRY INPUT=,OUTPUT=<>,SCRATCH= 2926 MOVAQ G^SCS$GQ_POLL,R1 ; get process polling block q head 2927 MOVL R1,R0 ; Save the Queue Head address 2928 10$: MOVL (R1),R1 ; Get Next SPPB 2929 CMPL R1,R0 ; Back to the Start? 2930 BEQL 20$ 2931 PUSHR #^M 2932 CMPC3 #16,L^MSCP_PROC_NAME,- ; Is this ours? 2933 SPPB$B_PROCNAM(R1) 2934 POPR #^M 2935 BNEQ 10$ ; 2936 MOVAL L^CNX$NEW_DISK_SYSTEM,- 2937 R4 2938 CMPL SPPB$L_RTN(R1),R4 ; Same Routine? 2939 BNEQ 10$ 2940 CLRL R0 ; Disable Polling 2941 JSB G^SCS$POLL_MODE ; Turn it off... 2942 20$: RSB 2943 2944 2945 2946 2947 .END ****** File WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 2794 .END ************ Number of difference sections found: 8 Number of difference records found: 155 DIFFERENCES /IGNORE=()/MERGED=1/OUTPUT=WORK8:[PERCIVAL.CLD.SYSLOA.SRC]DIFF.TXT;3- WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;22- WORK8:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 For VAX:- ************ File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;5 2 .IDENT 'X-24' 3 ; ****** File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 2 .IDENT 'X-23' 3 ; ************ ************ File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;5 41 ; X-24 JIP0011 Ian Percival 09-Jun-1997 42 ; Prevent unnecessary process polling overhead by removing 43 ; polls to disk controllers. 44 ; 1. Add new section to CNX$INIT to poll for MSCP$DISK 45 ; 2. Add new routine CNX$NEW_DISK_SYSTEM 46 ; 3. Add new routine CNX$STOP_DISK_POLL and calls to it. 47 ; 48 ; X-23 DAG Doug Gordon 23-Aug-1996 ****** File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 41 ; X-23 DAG Doug Gordon 23-Aug-1996 ************ ************ File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;5 432 $SPPBDEF ; SCS Process Polling Block 433 $TQEDEF ; TQE offsets ****** File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 425 $TQEDEF ; TQE offsets ************ ************ File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;5 470 MSCP_PROC_NAME: ; SYSAP name for MSCP server 471 .ASCII "MSCP$DISK" 472 .REPT 16-<.-MSCP_PROC_NAME> 473 .ASCII " " 474 .ENDR 475 476 CLU$GL_ACKLIM:: .LONG 0 ; Calculated ACKLIM ****** File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 462 CLU$GL_ACKLIM:: .LONG 0 ; Calculated ACKLIM ************ ************ File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;5 727 728 MOVAB W^CNX$NEW_DISK_SYSTEM,R0; Address of new disk system routine 729 ; R1 still contains the CONMAN SPPB 730 ; Store this in the new SPPB context 731 ; field for later use... 732 MOVAB L^MSCP_PROC_NAME,R2 ; Address of process name 733 JSB G^SCS$POLL_PROC ; Poll for copies of MSCP Server 734 CLRL R2 ; All systems at this time... 735 JSB G^SCS$POLL_MODE ; Enable polling 736 ; Do not worry about failure of the 737 ; disk polling - this will only leave 738 ; things the same as the Day 1 739 ; pre-Raven situation. 740 BSBW CNX$CON_INIT ; Initialize configuration manager ****** File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 713 BSBW CNX$CON_INIT ; Initialize configuration manager ************ ************ File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;5 766 ; R0 is context longword - unused by this sysap 767 ; R1 is address of process name ****** File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 739 ; R0 is context (nothing interesting) 740 ; R1 is address of process name ************ ************ File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;5 770 ; R4 is address of node name 771 ; ****** File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 743 ; ************ ************ File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;5 949 BSBW CNX$STOP_DISK_POLL ; Disbale polling this system for disks 950 POPR #^M ; Restore connection message addr ****** File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 921 POPR #^M ; Restore connection message addr ************ ************ File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;5 1400 BSBW CNX$STOP_DISK_POLL ; Disbale polling this system for disks 1401 MOVB #CSB$K_OPEN,- ; Mark connection open ****** File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 1371 MOVB #CSB$K_OPEN,- ; Mark connection open ************ ************ File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;5 2763 2764  2765 .SBTTL CNX$NEW_DISK_SYSTEM - Detect new disk controllers 2766 ;++ 2767 ; FUNCTIONAL DESCRIPTION: 2768 ; 2769 ; CNX$NEW_DISK_SYSTEM - Detects whether the node in question is 2770 ; a VMS node or is a disk controller. If the new system is solely 2771 ; a disk controller, polling is disabled for both the connection 2772 ; manager and the MSCP$DISK service. This reduces the anount of 2773 ; polling overhead the connection manager imposes. 2774 ; 2775 ; CALLING SEQUENCE: 2776 ; 2777 ; JSB CNX$NEW_DISK_SYSTEM - we are called here by the SCS Directory 2778 ; lookup services. 2779 ; 2780 ; INPUT PARAMETERS: 2781 ; 2782 ; R0 is context longword - we use it for storage of the CONMAN SPPB 2783 ; R1 is address of process name 2784 ; R2 is address of system ID 2785 ; R3 is address of process information 2786 ; R4 is address of node name 2787 ; 2788 ; 2789 ; IMPLICIT INPUTS: 2790 ; 2791 ; NONE 2792 ; 2793 ; OUTPUT PARAMETERS: 2794 ; NONE 2795 ; 2796 ; IMPLICIT OUTPUTS: 2797 ; SCS Polling will be disabled to non-OpenVMS nodes. 2798 ; 2799 ; COMPLETION CODES: 2800 ; 2801 ; R0 - 1 = disable polling to this system 2802 ; 0 = continue polling this system 2803 ; 2804 ; SIDE EFFECTS: 2805 ; 2806 ; 2807 ;-- 2808 CNX$NEW_DISK_SYSTEM: ; 2809 MOVL R0,R4 ; save the SPPB for CONMAN 2810 MOVL R2,R3 ; Save the System ID address 2811 MOVL R2,R1 ; Address of System ID 2812 CLRL R2 ; No output buffer 2813 JSB G^SCS$CONFIG_SYS ; Locate the System Block for this node 2814 BLBC R0,20$ ; R1 = SB on success 2815 CMPL SB$T_SWTYPE(R1),- ; Is this a VMS node? If so disbale 2816 #^A"VMS " ; polling ourselves.... 2817 BNEQ 10$ 2818 MOVZBL #SS$_NORMAL, R0 ; Disable polling to this node 2819 RSB ; CONMAN will form its own connection 2820 2821 10$: CLRL R0 ; Disable Polling 2822 MOVL R4,R1 ; SPPB 2823 MOVL R3,R2 ; System ID address 2824 JSB G^SCS$POLL_MODE ; Stop CONMAN polling a Disk Controller 2825 RSB ; Success disables poll for MCSP$DISK 2826 2827 20$: CLRL R0 ; Continue to poll on failure 2828 RSB 2829 2830 .SBTTL CNX$STOP_DISK_POLL - Stop looking for Disk Controllers 2831 ;++ 2832 ; FUNCTIONAL DESCRIPTION: 2833 ; 2834 ; CNX$STOP_DISK_POLL - This routine is called when a connection 2835 ; has been formed between this node and another's connection 2836 ; manager. This means that we have established the identity of the 2837 ; remote node - and so no longer need to poll for MSCP$DISK. 2838 ; 2839 ; CALLING SEQUENCE: 2840 ; 2841 ; JSB CNX$STOP_DISK_POLL 2842 ; 2843 ; INPUT PARAMETERS: 2844 ; 2845 ; R2 is address of system ID 2846 ; 2847 ; 2848 ; IMPLICIT INPUTS: 2849 ; NONE 2850 ; 2851 ; OUTPUT PARAMETERS: 2852 ; NONE 2853 ; 2854 ; IMPLICIT OUTPUTS: 2855 ; SCS Polling for MSCP$DISK will be disabled to remote nodes. 2856 ; 2857 ; COMPLETION CODES: 2858 ; NONE 2859 ; 2860 ; SIDE EFFECTS: 2861 ; 2862 ; 2863 ;-- 2864 CNX$STOP_DISK_POLL: ; 2865 PUSHR #^M 2866 MOVAQ G^SCS$GQ_POLL,R1 ; get process polling block q head 2867 MOVL R1,R0 ; Save the Queue Head address 2868 10$: MOVL (R1),R1 ; Get Next SPPB 2869 CMPL R1,R0 ; Back to the Start? 2870 BEQL 20$ 2871 PUSHR #^M 2872 CMPC3 #16,L^MSCP_PROC_NAME,- ; Is this ours? 2873 SPPB$B_PROCNAM(R1) 2874 POPR #^M 2875 BNEQ 10$ ; 2876 MOVAL L^CNX$NEW_DISK_SYSTEM,- 2877 R4 2878 CMPL SPPB$L_RTN(R1),R4 ; Same Routine? 2879 BNEQ 10$ 2880 CLRL R0 ; Disable Polling 2881 JSB G^SCS$POLL_MODE ; Turn it off... 2882 20$: POPR #^M 2883 RSB 2884 2885 2886 2887 2888 .END ****** File WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1 2733 .END ************ Number of difference sections found: 10 Number of difference records found: 157 DIFFERENCES /IGNORE=()/MERGED=1/OUTPUT=WORK302:[PERCIVAL.DOCS]DIFF.TXT;18- WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.MAR;5- WORK302:[PERCIVAL.CLD.SYSLOA.SRC]CNXMAN.ORIGINAL;1