Computer Acceleration Record

PowerAnts

Computer Acceleration Record [Copy link]

Office and commercial PC: i5-4670, 8G DDR3, 500G 64M 7200 rpm HD (I hate this kind of junk hard drive the most), it is stuck. The network administrator refused to add an SSD, I couldn't stand it anymore, so I started my own business. As we all know, the bottleneck of this configuration is the HD, but the problem of HD itself is concentrated on 4K reading and writing. The minimum unit of external storage addressing of a computer is 4KB, and files smaller than 4KB also occupy 4KB. Some abnormal software consists of a large number of KB files, such as QQ, Wangwang, IE cache, system TMP files, etc. In fact, single-threaded software like POWERPCB consists of a few large files, and the speed increase effect of SSD is not very obvious. If you keep reading 500MB/S of data, the CPU can't keep up. SATA3 HD is just right, and it is actually QQ, Wangwang, and the browser that are stuck... The maximum addressing time of HD is 6mS, and most USB flash drives only have 0.5mS, which means that when reading a large number of small files, the performance of USB flash drives will be much better than HD. It would be nice if large files were read from HD and small files were read from USB. Through understanding, Windows' readyboost can use USB as an accelerator for HD. After enabling readyboost, some small files that need to be read frequently will be thrown into the USB by the system and a list will be generated. When these small files need to be read again in the future, they will be directed to the USB, thereby improving the access speed of small files. This thing actually creates a file called readyboost.sfcache on the USB. The function of this file is actually to occupy the pit and lock the data storage location to avoid being misappropriated by FAT. The system encapsulates the Prefetch pre-read file into the readyboost.sfcache file, which is equivalent to bypassing FAT for direct addressing, and does not change its size, date and other attributes. The readyboost.sfcache file has a mirror image on the system disk. When the USB is removed, the system will automatically switch to the mirror file for access, but the 4K read and write speed will become HD level, and everything will return to the past without any adverse reactions. When Microsoft first launched readyboost, it promised that the life of the USB used for readyboost would be more than 10 years. It is not clear whether wear leveling is done, but from the working principle of this application, it should be a read-only buffer, not a write buffer. In addition, the worst early USB flash drives were MLC, with a write life of more than 10,000 times. As a read-only data block, it basically has no significant impact on the life. I have several USB flash drives on hand. The 32G and 64G TOSHIBA USB flash drives with USB3.0 purchased in recent years are actually TLC particles, with a write life of only 500 times. I am still a little worried about using them. However, a 4G USB2.0 USB flash drive that has been used for more than 10 years is actually SLC. After plugging it in and enabling readyboost, the USB flash drive light flashed for almost ten minutes, and it was writing crazily. After stopping, I tested the 4K read rate, which jumped from 0.6M/S in HD mode to 3M/S in HD+U disk mode. The write rate was 0.8M/S with no change (HD's 4K random write is slightly better than random read, which is due to the assistance of large-capacity cache. When writing, the system only needs to send data to the hard disk buffer. However, DRAM has a refresh cycle, which is 64mS at most. If you don't want to waste resources to refresh, you must try to write the data in the buffer to the specified sector in advance. It is not clear whether there is a reply after writing. I didn't study it too much. I was secretly happy for a few minutes. The data looks good, but after two days of use, it seems to be useless. Next, use primoramdisk to divide 2G into ramdisk, make a mirror backup before shutting down, and re-import it when booting up. The acceleration effect is very good, but ramdisk is most afraid of power outages. If there is a power outage, it will be terrible. Forget it. Moreover, it occupies 2G of memory, and it is very tight to open a virtual machine, so this method is also invalid. Finally, lock primocache, format the 4G SLC disk above into binary storage format, turn the USB flash drive into a directly addressed storage device, bypass the file system and become a secondary cache between the system memory and the HD. However, I only enabled the read buffer and half-closed the write buffer. That is to say, only the data read from the disk will enter the USB flash drive. The data that the operating system needs to write to the HD will not be written to the USB flash drive, so as to avoid wear and tear on the USB flash drive. How is the result? Please see Take a look at the program folder of Aliwangwang. It is so abnormal, with an average size of 16K. Assuming that half of the startup program needs to be read out, and assuming that the 4K blocks of these 16K files are continuous, then the HD will take at least 5 seconds to read them all out. In fact, the average reading time of the HD will double, plus user files, C++ libraries, etc., no wonder it takes more than 20 seconds. These large numbers of small files are cached on the USB flash drive and hit in a directly addressed manner, which only takes three seconds. No wonder it can be opened in seconds. Excel feels the best

maychang

This is quite useful.

PowerAnts

maychang posted on 2018-9-16 09:59 This is quite useful.

The yellow line with the largest fluctuation is the disk read rate curve of the 2TB notebook mechanical hard disk with 128M cache that I bought ten days ago. The other colors are the speeds of 11 USB flash drives with 1G-64G on hand on the USB2.0 port (the oldest one is at least ten years old). It can be seen that even with USB2.0, the USB flash drive is still 5-10 times faster than the mechanical disk in reading files below 64KB, and this advantage is maintained up to 128-512KB. In fact, similar to HD, the interface rate of USB flash drives can only improve large files. The IOPS of small files is basically determined by the addressing speed. The effect of USB3.0 may not be obvious. Although the SATA3 interface rate of HD reaches 6G/S, it can actually reach 550-600MB/S after deducting the encoding. In fact, there are 746 heads in one circle of the outer ring of HD. At a speed of 7200 rpm, the theoretical upper limit of the continuous reading rate of a single-disk double-sided is 2*512*746*120/1024/1024=87.42 (MB). Too high an interface rate does not help much for reading, but it helps a lot for writing. According to the previous data, the transmission time during writing can be compressed by about 7 times. For example, with a 128M cache (if the cache rate can fully keep up with SATA3), it will only take 232mS for the motherboard to write 128M of data to the hard disk, while it will take 1464mS to actually write to the disk.

PowerAnts

Correction: Actually, there are 746 heads in one circle of the HDD. It should be 746 sectors.

懒猫爱飞

It looks very exciting, but please forgive me, a computer layman, I don’t understand what you are talking about. I only know roughly that you are a computer expert and seem to have made the computer faster:titter:

wshqna

This is for mechanical hard disks, right? Does solid state drive have this bottleneck? I don't quite understand, hope you can give me some advice! !