Following are the OS Tunings only for all the Servers and clients, not any Installation steps:
*set maxusers=2048: Number of concurrent users on the system determining the amount
of physical memory allocated to the kernel
*set maxphys=8388608: Hard limit on the open file descriptors for a single process
*set rlim_fd_max=1048576: Hard limit on the open file descriptors for a single process
*set autoup=600: Controls the frequency in seconds of entire physical memory to
be scanned for dirty pages
*set tune_t_fsflusher=1: Reduces fsflush daemon overhead
*set ufs:ufs_HW=1572864: Limit the amount of I/O that may be outstanding
to a single file on a system-wide basis
*set ufs_LW=1048576: When ufs_HW bytes are outstanding, I/O requests will
sleep until less than ufs_LW bytes are outstanding
*set sq_max_size=512: The depth of the syncq (number of messages) before a
destination streams queue generates a QFULL
*set segmap_percent=64: Amount of memory used, to hold file system pages
*set shmsys:shminfo_shmmax=0xffffffffffffffff: Maximum size in bytes of a shared memory segment
*set shmsys:shminfo_shmseg=512: Maximum number of semaphores in the system
*set shmsys:shminfo_shmmni=4096: Systemwide maximum number of semaphore undo structures
*set semsys:seminfo_semmsl=340: Maximum number of semaphores per semaphore set
*set semsys:seminfo_semmni=320: Establishes the maximum number of systemwide semaphore sets
*set msgsys:msginfo_msgmax=16384: Maximum size a message can be, in bytes
*set msgsys:msginfo_msgmnb=16384: Maximum number of bytes on a message queue
*set msgsys:msginfo_msgmni=2200: Number of message queue identifier
*set msgsys:msginfo_msgtql=2500: Maximum number of message headers
*set msgsys:msginfo_msgseg=32767: The number of units of space that can be allocated
from the memory map segment
*set msgsys:msginfo_msgssz=128: Maximum message segment size
*set msgsys:msginfo_msgmap=200: Number of entries in the message map
Following are the NETWORK Configuration and Tunings:
*One Asente, GX5-1400 Gigabit Workgroup Switch is used for the network and routing
*All 4 clients are connected to the switch via 1 Sun[TM] Gigaswift Ethernet on board controller/client, 1Gbit/s
*2 POPSTORE servers and the MMP server are connected to the switch via 1 Sun[TM] Gigaswift Ethernet
on board controller/server, 1Gbit/s
*The SMTPMTA server and the LDAP server are connected to the switch via 1 Sun[TM] Gigaswift Ethernet external
controller/server, 1Gbit/s
*All the network connections and the routing are resolved using UNIX hosts file
*set ce:ce_bcopy_thresh=96: The mblk size threshold used to decode when to copy a mblk
into a pre-mapped buffer, or use DMA methods
*set ce:ce_taskq_disable=1: Disables the use of Task queues and forces all packets to go
up to layer 3 in the interrupt contest
*set ce:dvma_thresh=95: The mblk size threshold used to decide when to use the fast path
DVMA interface to transmit mblk
*set ce:ce_reclaim_pending=16: This is the threshold for when reclaims start happening
*set ce:ce_tx_ring_size=8192: This value is the size of each Tx descriptor
*set tcp:tcp_conn_hash_size=262144
*/usr/sbin/ndd -set /dev/tcp tcp_recv_hiwat 65536
*/usr/sbin/ndd -set /dev/tcp tcp_xmit_hiwat 65536
*/usr/sbin/ndd -set /dev/tcp tcp_conn_req_max_q 16384
*/usr/sbin/ndd -set /dev/tcp tcp_conn_req_max_q0 16384
*/usr/sbin/ndd -set /dev/tcp tcp_smallest_anon_port 8192
*/usr/sbin/ndd -set /dev/ip ip_ignore_redirect 1
*/usr/sbin/ndd -set /dev/tcp tcp_keepalive_interval 30000
*/usr/sbin/ndd -set /dev/tcp tcp_naglim_def 1
*/usr/sbin/ndd -set /dev/tcp tcp_mss_min 108
*/usr/sbin/ndd -set /dev/ip ip_respond_to_echo_broadcast 0
*/usr/sbin/ndd -set /dev/ip ip_forward_directed_broadcasts 0
*/usr/sbin/ndd -set /dev/tcp tcp_time_wait_interval 60000
*/usr/sbin/ndd -set /dev/ce instance 1
*/usr/sbin/ndd -set /dev/ce rx_intr_time 500
*/usr/sbin/ndd -set /dev/ce rx_intr_pkts 24
*MSL is 30 seconds, TIME_WAIT is 60 seconds, MTU is 1500 bytes (equates to MSS 1460 bytes): Fixed parameters
Reference: Sun Java System Messaging Server 6.1, Tuning Guide: http://docs.sun.com
The benchmark was conducted at the Sun's Performance lab in Burlington, Mass
|