System tuning and vm_bigpg_enabled

From: LAWRIE SMITH (lawries@btinternet.com)
Date: Thu Apr 22 2004 - 05:21:56 EDT


Admins,

I am trying to peformance tune a 4 node V5.1b-1
cluster of 2 x ES40 (8Gb RAM, 4 x 500Mhz CPU) and 2 x
ES45 (8Gb RAM, 2 x 1Ghz CPU) running multiple
instances of Progress RDMS applications. The
sysconfigtab is listed at the end of this mail.

Any advice would be welcomed but specifically, I am
trying to find out more about the vm_bigpg parameters.
I have read the documentation but remain unsure of the
significance of these parameters and where it would be
appropriate to use them, and how to discover this.

Can anyone explain this stuff to me, or point me at a
white paper etc? I see nothing about these parameters
in the tuning manuals.

Thanks in advance (will summerise)

  vm_bigpg_enabled
      The master switch that enables (1) or disables
(0) memory allocation
      for user processes in "big pages" mode.

      Default value: 0 (disabled)

      Big pages memory allocation allows a virtual
page in the process
      address space to be mapped to multiple pages in
the system's physical
      memory. This mapping can be to 8-pages,
64-pages, or 512-pages (64,
      512, or 4096 Kbytes, respectively) of physical
memory.

      Big pages uses threshold values set on a per
memory-type basis to
      determine whether a memory allocation request is
eligible to use the
      extended page sizes. The attributes that set
these thresholds are
      vm_bigpg_anon, vm_bigpg_shm, vm_bigpg_ssm,
vm_bigpg_seg, and
      vm_bigpg_stack.

      If big pages memory allocation is disabled, the
kernel maps each vir-
      tual page in the user address space to 8 Kbytes
of memory.

      To enable big pages, you must set the
vm_bigpg_enabled attribute at
      system boot time.

generic:
 memberid=1
 msgbuf_size=524288
 new_vers_high=1445673072572143232
 new_vers_low=268488508
 act_vers_high=1445673072572143232
 act_vers_low=268488508
 rolls_ver_lookup=0
 kmem_debug=0x40
 kmem_protected_size=32
 version_vendor = Compaq Computer Corporation
 version_avendor = COMPAQ
 version_product = Tru64 UNIX
 version_banner = Compaq Tru64 UNIX

vm:
 swapdevice=/dev/disk/dsk0f,/dev/disk/dsk3f
 vm_page_free_reserved=20
 vm_page_free_min=30
 vm-swap-eager = 1
 vm-page-free-target = 2048
 vm-ubcseqstartpercent = 50
 ubc_maxpercent = 70
 vm_ubcdirtypercent=30
        ubc_minpercent=10
 vm_page_prewrite_target = 5500
 vm_segmentation = 1

proc:
 max_proc_per_user=6000
 max_threads_per_user=4096
 round_robin_switch_rate=60
 autonice=1
 autonice_time=300
 autonice_penalty=15
 per_proc_data_size=10737418240
 max_per_proc_data_size=10737418240
 max_per_proc_address_space=4398046511104
 per_proc_address_space=4398046511104

inet:
 udp_ttl=60
 udp_recvspace=262144
 tcp_keepalive_default=1
 tcp_keepidle=3600
 ipport_userreserved=65000
 ipqmaxlen=2048

socket:
 sominconn=65000
 somaxconn=65000

pts:
 nptys=512

vfs:
 bufcache = 1
 noadd_exec_access = 0

ipc:
 shm_max=1073741824
 shm_seg=512
 sem_mni=2048
 sem_msl=2048
 shm_mni=2048
 sem_opm=400
 sem_ume=400

sec:
 acl_mode = enable

lsm:
 lsm_rootdev_is_volume=0

clubase:
 cluster_expected_votes=5
 cluster_name=wm-hart-cluster
 cluster_node_name=CH004
 cluster_node_inter_name=CH004-ics0
 cluster_node_votes=1
 cluster_interconnect=mct
 cluster_seqdisk_major=19
 cluster_seqdisk_minor=143
 cluster_qdisk_major=19
 cluster_qdisk_minor=383
 cluster_qdisk_votes=1

advfs:
        AdvfsAccessMaxPercent=35
        AdvfsSyncMmapPages=0
cfs:
        cfsiosize=131072

Lawrie



This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:49:57 EDT