How to choose hardware
Internet mainstream hardware composition:
cpu: 32porcessor (8 physical cores, 2 processors per core, hyperthreading enabled) 2.5GHZ
Memory: 32G>64G>96G>128G
Hardware: SATA mechanical disk>SAS mechanical disk>SSD solid state drive (300M--500M)
10 times the price, 50 times the IO performance, faster and faster read and write speeds, and higher hardware costs
Network card: 100mbs 1000mbs
linux cat /proc/couinfo to check the cpu status
free -g View memory situation -g is in g
Classification of hardware:
Model classification :
In-memory (cache cluster)
CPU：8CoreX2 Mem:128G Disk:SAS 600G*6 RAID5
CPU：8CoreX2 Mem:192G Disk:SAS 600G*6 RAID5
I / O type (DB)
CPU：8CoreX2 Mem:128G Disk:SAS 600G*2 RAID1+Intel S3700 800GX6 RAID5
Storage-Sprk
CPU：8CoreX2 Mem:192G Disk:SAS 600G*2 RAID1+SATA 4TBX12 Non-RAID
Storage-Hadoop
CPU：8CoreX2 Mem:128G Disk:SAS 600G*2 RAID1+SATA 4TBX12 Non-RAID
storage-public
CPU：8CoreX2 Mem:192G Disk:SAS 600G*2 RAID1+SAS 1TBX24 Non-RAID/RAID5
Computational (special machines)
CPU：10CoreX4 Mem:192G Disk:SAS 600G*6 RAID5
CPU: 8CoreX2 Mem: 192G Disk: SAS 600G*6 RAID5 GPU card graphics comparison
58 server model list
R410 mainstream server, currently used the most
R420 mainstream server, currently used the most
C6100 2U server is equivalent to 4 independent hosts
R710 mainstream server, DB database server is the most used
R720 DB database server is the most used
Log storage for R515 op
R720xd DB database hadoop server
The concept of U: server length, width and height interface specification, server unit.
Choosing Hardware - Hardware selection depends on the business use case
Web Business Scenario----Memory/Computation
Logical Business Scenario - Computational
Cache application----memory type
Test application - virtual machine
Database scenario----DB storage type, SSD as much as possible
Real-time Computing Scenario----Storage-Sparks/Storm
Offline computing scenarios - storage-type Hadoop requires high disk capacity
Mass data storage ---- files, pictures, etc. If the storage type provides online services, it needs to provide cpu power.
Image Recognition Computing Scenario ---- Computational GPU
Offline, edge business scenarios - virtual machines, physical isolation
Traditional Enterprise Configuration
IOE enterprise software and hardware configuration ---> IBM server ORACLE software MEC storage device
Internet business configuration
PC-level servers (low price), cheap equipment, and low-priced PC servers are down once a year, which is a probabilistic event. High-intensity and frequent reading and writing of ordinary hard drives, the probability of damage is higher. Hardware availability is further reduced.
How to achieve high availability at the hardware level?
Try to use high-profile servers. If not, focus on high disk availability: hard disk specifications SATAIII SCSI SAS SSD, in this order, the performance is getting better and better, the price is getting higher and higher, and the availability is getting higher and higher. Use high-availability disks as much as possible for redundancy. RAID disk array. RAID0 RAID1 RAID5 RAID10. RAID optimization: stripeElementSize of this element size, DB is set to 128K, and the general application is set to 64K. The read-write strategy fore web with no battery does not apply to forced write-back by battery.
For multi-machine redundancy, at least 2 nodes are used in the module application, and how much is required depends on the overall system throughput and single-machine throughput.
Storage application, Mysql through Master-Salver. Mango Replic-Sets, the master node can be automatically selected after the master node dies.
Think, hypothetical scenario:
1000w records, each record is 1K, MySql storage, using SSD hard disk, the read and write throughput is 3KQPS/TPS, and the read-write ratio is 10:1. Can the database withstand this scenario, and is it necessary to add caching?
The read and write data of the SSD is 300M per second, and the mysql software can reach 200M per second, 3KQPS, 3M for each request for 1 piece of data, 30M for each request for 10 pieces of data, and 300M for each request for 100 pieces of data. At this time, the system does not meet the requirements and needs to use the cache.
The relationship between Taobao's TPS and PV is usually the highest TPS: PV is about 1 : 11*3600 (equivalent to 11 hours of access at the highest TPS, this is the scene of product details, and different application scenarios will be slightly different)