RH442 1

标签：java   os   io   数据   for   art   cti   ar   

hardware profiling

RH442

Getting to know your hardware

Hardware profiling

Before tuning the performance, it‘s important to understand the capabilities of your system
    CPU 内部寄存器:硬件,时钟周期一致.1级缓存:静态内存,2级缓存:动态高速内存
    Memory 相对缓存更慢 
    Storage 更慢 到磁盘，终止。
Farthe from CPU,slower,bigger,less frequently accessed
Three elements of X86 machine, the ways to use system 针对X86结构
    I/O address
    IRQ
    DMA

CPU profile 指标

grep CPU /proc/cpuinfo

model name : Intel® Xeon® CPU E5-2620 v2 @ 2.10GHz

model name : Intel® Xeon® CPU E5-2620 v2 @ 2.10GHz

Model,Hz 操作系统认为你有两个cpu,Hz是跑步的频率，步子多大无关系
Caches size, L1, L2, L3, private or shared

lscpu

L1d cache:32k L1缓存分为数据缓存

L1i cache:32k &指令缓存

L2 cache:3072K

cpu(s):4 cpu数量为1

Thread(s) per core:1 每颗物理CPU，1为无超线程

Core(s) per socket:2 每颗物理CPU有两个核心

Socket(s):2 两颗物理CPU

CPU features-VT,Architecture 虚拟化,架构64/32

是否支持64 cat /proc/cpuinfo |egrep ‘(lm)|(pae)‘

Motherboard chipset 主板芯片组，决定主板和CPU效率

No.of sockets,cores,hyper thereading support

Interconnect between sockets 物理CPU多颗时。前端总线联系FSB/qpi/amd 物理CPU之间通讯手段

TOP CPUs

intel E7-8830
8 cores,16 threads 8核，16线程
24M cache, 2.13 GHz,QPI

AMD 6284 SE
16 cores
16M cache 2.7GHz

Memory profile 内存

Memory size, max allowed
> free -m

>cat /proc/meminfo

Memory technology,bandwidth and latency 内存第几代技术，以及带宽延时

>DDR2(2 bits),DDR3(4bits),DDR4(8 bits),Rambus(pS/2)

>Latency(wait time before read agin,in ns) 反映速度 纳秒级 列刷新频率

>ECC 内存故障修复 发现内存位失效 纠正 减速 slower safer

Method of memory accessing- UMA NUMA

    频率/2=速率 2133/2~~ 1G

UMA or SMP

>MCH 内存控制器 速度瓶颈  pcieX16 显卡占用 卡死了 这是北桥

>ICH 南桥连外设 sata usb pcie*1

NUMA 架构

>两个CPU自己管理自己内存，MCH内置到CPU

>IOH 伪北桥，IO控制器

>ICH 南桥一样

NUMA form apple Mac Pro

>MCH & ICH ->I/O HUB 内设& I/O CONTROLLER 外设..GPU很近

Storage Profile

Type of storage used
> Mechanical magnetic platters 盘片 ，磁碟

> SSD devices

Hardware RAID level, no. of disks, striping size

Interconnect type for direct attached storage 内部存储
>SATA SAS IDE    IDE没人用了 sata企业级,第三代sata完虐sas sas可靠

SCSI,FIBRE,CHANNEL,ISCSI PCI总线接驳，光纤，互联网

> 带宽，纳秒级，便宜

burstspped vs. average speed 顺序读写 随机读写

SSD

Electronic disk , no moving mechanical components

No start-up time

Very low latency on read and write

Faster than traditional HD sequence 500M/S 200M/S

Silent ,not so hot ,vibration free, magnet free,light,less power required

Much more expensive,lesser capacity

Potentially longer life time


MLC 一个单元存多位闪的多寿命短
SLC 很贵，非常贵

每次读至少一个block

####SSD TRIM COMMAND

when os erases a file, the file will typically be marked for deletion
SSD 不停做垃圾回收
ssd 挂载 fstab default,discard


>IDE UP 133MB/S

>SATA2 300

>SATA3 600

>SCSI 320

>USB2 60

>USB3 600

>ISCSI GIGA NIC 125

>FIBRE CHANNEL 531


###PCI EXPRESS

>1.0a 2Gb/s bandwidth:2.5GT/s singleLane:250Mb/s full16:8GB/s

>2.1 4      5       500     16

>3   8      8       1GB/s       32

Networking profile

Network card model
Network technology
Connection to switch(Speed)
Separated network, special purpose network -Storage,VDI,administration,useradta
I/O virtualization

>cmd

    >dmesg

    >/var/log/dmesg

    >lscpu,lscpu -p

    >getconf -a

    >x86info

profiling commands

>cmd

    >dmidecode

    >/sys/class/dmi

    >lspci [-v[-v[-v]]]

    >lsusb [-v]

    >ethtool

    >sosreport

    >powertop

RH442 1,布布扣,bubuko.com

RH442 1

标签：java   os   io   数据   for   art   cti   ar   

原文地址：http://my.oschina.net/HankCN/blog/297499