| Back to logs list
Reprinted from 228191261 at 20:29 on January 7, 2011 Reading (loading. ..) Comments (0) Category: Personal Diary
SSD for storage SSD is available, is undoubtedly a revolution, destined to become the destroyer of traditional hard drives before! the current situation because of many constraints, SSD hard drive has not been fully universal; and SSD hard drives of many advantage is exactly suited to the characteristics of the diskless server applications, as a focus on the diskless server hardware product development, and has been committed to bring customers the best of the diskless server hardware solutions for the professional company - Guangzhou Rui Chun (INTEL SSD Solid State full application of the hard disk read and write in two aspects of the diskless server, the actual operating results last for six months tells us that with the right combinations of technology solutions, SSD hard drives on the full application of the diskless no longer a distant and long, you can SSD can be said that the current round has been completely diskless servers in the application; performance naturally without doubt, life will never SSD vulnerable areas, we are pleased to see the full adoption of diskless servers INTEL SSD greatly improved with the machine volume, while prices fell a lot, and also save a lot of power consumption! Through this analysis of the application of SSD diskless, hoping to pay more attention to friends of SSD diskless diskless technology applications have a more in-depth understanding of .
one, SSD Solid State Disk features
features introduced on the SSD, we have to two types of SSD storage unit, and with the traditional mechanical hard disk overview of the working principle of contrast.
storage unit is divided into two types: SLC (Single Layer Cell single-level cell) and MLC (Multi-Level Cell MLC). SLC is characterized by high cost, small capacity, fast, and MLC is characterized by large capacity, low cost, write speed and IOPS than SLC. Each unit is 2bit MLC, relatively SLC is a whole more than doubled. However, the MLC memory cell in each store more data,
Sell your car!, the structure is relatively complex, the chance of error increases, the need for bug fixes, this action led to significantly lag behind the performance of the simple structure of SLC flash memory. In addition, SLC flash memory has the advantage of replication times of up to 100,000 times, 10 times higher than MLC flash. In addition, in order to ensure the life of MLC, the control chips and smart wear leveling technology checksum algorithm, so that each memory cell can be written to the average number of share, reaching one million hours time between failure (MTBF). MLC and SLC superior performance, is sufficient to meet the needs of SSD diskless applications, considering the cost, we will try to choose at the price advantage of the MLC SSD.
traditional hard drives are generally motor, disks and heads and other mechanical moving parts composition, the use of moving parts will require more time and greater power consumption, which is more than traditional hard drives in the diskless server, IO, and more reading and writing state operation, the client starts slow speed and one of the reasons to enter the game, and solid-state drive technology is the use of solid-state flash memory chips to replace the traditional mechanical hard drive moving parts, eliminating the need to wait for the movement of mechanical parts, thus greatly speeding up hard disk read and write data transfer speed, whether it is a diskless server disk in high-load, full machine load conditions, the client without waiting for server disk I / O processing speed problems remain.
1, the performance is far higher than the mechanical hard disk
MLC single disk read speed, standard 250MB / S, write speed of 100MB / S, the SLC's write speed is more than 200MB / S, 2-3 times higher than the traditional hard drive! seek 0.1MS, existing conventional 7200 RPM enterprise-class drives, seek time is 12MS, and 15,000 rpm SAS hard drives, also located in 5.6MS so, the key indicator is the IOPS that the parameters of I / O per second, per second read and write (I / O) the number of operations, and more for the database and so on, to measure the performance of random access. SSD Solid State Drive in the 64KB block random read standard, up to 2000 can handle multiple tasks, and 15,000 rpm SAS hard drives can only handle 170, can be seen, SSD in a diskless application on the market than the mainstream SAS hard disk is more advantageous to break the bottleneck of the traditional hard drive read and write speeds.
2, low power consumption
SSD standby power consumption is only 0.05-0.1W, compared to traditional hard drives ranging from 5-10W (depending on manufacturers to design and drive type), while the operating power consumption, SSD is only 2-4W! in our practice tests in a diskless server SSD, 1 hour of power is 0.07Kwh, traditional diskless server disk power consumption is 1 hour 0.13Kwh.
3, low heat
mechanical arm device, and thanks to exempt low-power design, SSD's heat is quite low, and traditional hard disk is 40 degrees or more frequently if the diskless server cooling structure is not very reasonable in the circumstances, heat system fan growth over time lead to performance degradation, environmental range of issues,
new new balance shoes, to promote the hard disk temperature is too high, life expectancy reduced by half, resulting in the loss of customers is self-evident, and cooling advantages of SSD, the time will become a customer purchases highlight!
4, no noise
SSD and traditional hard disk completely different, completely 0 noise operation. The traditional hard drive noise is to a considerable number of diskless elite headaches.
5, light weight
General SSD weighs about 90 grams, a 3.5-inch SAS hard disk normal weight is around 700 grams, from the point of packaging, logistics and cost considerations, SSD solid state hard drive than the traditional hard drive adds an advantage!
6, seismic intensity
disk in the logistics process, or in the installation process, if the case had any serious falls, while in the hard disk into a dead state, ranging from the hard drive hidden deep buried,
new balance outlet! and traditional hard disk work if there is vibration, it is very easily lead to
Second, SSD and diskless server applications
diskless workstation server machine running speed, and their own server maintenance costs have been is not a good solution, although various diskless software systems do a good optimization, but with the file system data volume increase, no matter how good it is difficult to solve the optimization of the machine's own I / O processing speed, as a diskless server's overall performance bottleneck is disk I / O processing speed is limited, although the disk array can be used to improve I / O processing speed, but this increased capacity is limited, and also increase the hardware cost, once the hard disk broken, or aging problems, but also the stable operation of the system. In today's fast development of computer technology, SSD technology has become more mature state, is undoubtedly the late spring of diskless industry!
the cost side, a SSD can serve as the most mainstream of three SAS hard drive array task, full use of all the serial program, reduce the use of integrated SAS chip, independent SAS card, the extra costs, and case more disk arrays, RAID 0 mode which increases the risk, if a disk fails after the data is all lost! can be found from a series of advantages, SSD application performance in the diskless service is completely beyond the traditional mechanical hard drives, lower cost, fully justified replace the existing disk configuration diskless servers, diskless server disk as a new application needs!
Third, SSD and the most mainstream of SAS hard drives comparison test
introduced to the medium-length article, we focus on analyzing the difference between SSD and traditional hard disk number, how much real value to give the diskless server!
following figure using Intel SSD MLC160 and most mainstream market, Seagate 300G 15K7 SAS hard to do comparison tests, the sustained read speed has been slightly better had.
then used the Intel SSD 64G SLC and transfer SAS Fujitsu 73G 15000 write speed comparison chart, SSD SLC has two FST 73G RAID 0 speed.
, of course, from the above test results, and not entirely representative of SAS SSD performance than a huge difference, MLC160G continuing to write the test of time, or less than ST15K7300GSAS, but this in itself is not a problem, in practical applications , continued to write the condition is not frequently occur, and the diskless server to write the application of plate is needed is a small file random disorder scattered write, continuous velocity is not important, more important is the need in the short time large number of small files quickly process, we then run, more important indicators of the diskless server disk performance tests, that is, the disk IOPS performance test!
diskless server data applications to 64K of data for the standard case, Intel SSD MLC 160G SAS disks have is 10 times the mainstream! This is because the characteristics of SSD with flash memory instead of machinery, will undoubtedly find the disk Road, the most important time to zero milliseconds index increased less, the test looks at the diskless application server's disk write to complete the absolute leader in terms of performance SSD mechanical hard drive.
Then we discuss an issue, diskless software to read and write separation technique used, the system image packages / mirror where all the game partition, the partition can be seen as IMG is the image and write data back to the client to the server separately from a different partition (array) on. This read / write separation,
new balance sneakers, the pressure on the server hard drive is relatively small, thereby enhancing the speed of the client, which is the conventional norms diskless server disk system is one, why not use a RAID hard drive or a group of all competent task, because a single disk or a RAID, read and write while the case of a serious decline in the performance of the disk, SSD diskless on the application streaming technologies also need to read and write it? let us have a look real test it.
on the map is an ST SAS 300G 15K7 SAS sustained read and write operations while the test read, seen from above, disk speed has 7MB / S is about, and this is why the do diskless reason behind the separation of reading and writing !
Then we
Intel SSD MLC 80G test, reading and writing while continuing to operate, disk speed, still close to 60MB / S rate is 10 times the mainstream of SAS hard drives, which read and write is not enough to show that the separation has been become a reality!
several properties from the above tests, SSD Solid State Drive performance has been far ahead of the traditional mechanical hard drives, full application of the diskless server just around the corner!
four, SSD use in the life of diskless
;
SSD life must have been the subject of particular concern to us, and in the application of diskless, whether their life is guaranteed! this, we discuss the working principle of SSD, and MLC and SLC SSD performance and reliability The difference, in fact, Intel for the SSD in the reliability and service life, there are some safeguards.
write mechanism and the intelligent wear double balanced
in-depth analysis SSD MLC and SLC's performance difference, we describe the working principle of the SSD. SSD is the smallest unit of record Page (4K), the minimum erase unit is the block (512K). When we save the file, it will be 2 pages in the form of memory stored in the flash memory; be removed, SSD will mark pages as invalid, but do not actually remove the block. SSD will wait until the block in a certain proportion of pages are marked as invalid, a number of valid data will be copied to the new pages to go and erase the block. SSD This is done to limit the number of times each block erase, thus extending drive life.
not handle all of the SSD removed through the same way, how to delete and when to remove some of the block contains an invalid page and become a key life of SSD.
written magnification
SSD for some poor quality, if you just want to replace a 16KB file, the controller may incorrectly read the entire block to main memory,
Second Valentine's Day - Qzone log, and then replace the four pages, erase from the SSD block, and then change the page 4 to the new block is written. Intel SSDs page / block, it means that if only the new data written to 16KB, it is equivalent to 512KB written SSD - written magnified 32 times.
we have no other SSD controller's data can show how they deal with this situation, but Intel said the traditional SSD controller written in 2040 times magnification range, which reduces their drive life.
zoom control needs to write cache
This is the Intel SSD's controller needs the reasons for DRAM cache, it uses as a temporary data storage memory, so more efficient than conventional flash memory controller to write the implementation of action; conventional flash memory each time the controller is usually are triggered by write action, even when stored in relatively small blocks will be written to zoom to a larger area.
inherent design advantages
Intel considers the average write amplification and wear and tear factor, a simple equation of NAND flash memory cycle count, the average wear factor against you wear and the maximum average utilization rate of wear differences, Intel claims that, due to the perceived flash controller memory block size is enabled with the command sequence features, so factors other products 3 (3x), X25-M is only 1.1x; This means that the utilization of all the storage room is almost equal, this will help to improve the expected SSD life.
workload for the typical customer, its write amplification factor of less than 1.1 times, in other words, more than you actually need to write less than 10% of the data. Write amplification factor itself does not make much sense, it is important the life of the drive, in addition to another factor at play, we have confirmed, for the flash memory, the limited number of write block, after which it will lose the ability to store data . SSD is smart, will use the balanced loss algorithm block scattered throughout the drive to use. Remember, the SSD is, no matter where written, its performance will be consistent, which is different from other mechanical disks, the efficiency of Intel wear a balanced, all using the block is almost the same value.
Therefore, SDD attempts to write data evenly all the block in the drive. For example, suppose you download a 2MB file to your new, never used the SSD, and then save it to block 10,11,12,
new balance shoes,13 in. You realize that downloading the wrong file and delete it, then go to download the correct file.
flash memory controller is not writing the new file block 10, 11,12,13, but instead of writing block 14,15,16,17. In fact, block 10,11,12,13 four block can not be used again until the drive every other block has been written once. So, although you may only have 10,000 MLC SSD life cycle, but the loss of the smart balance algorithm, it can use for a long period of time, wear bad balance, probably in the existing SSD, some blocks than other blocks more be used.
Intel SSD will be about 4% of the space is not a balanced loss of control algorithms, which means that Intel SSD 4% of the block on the block easier to wear than the other. Written by low magnification and the average wear of stable, mature technology, a fundamental extension of the hardware life!
SSD life often can be quantified, can not continue to use a block in before you can be erased through their design cycle times or to measure, because we are in front of MLC is generally 10,000 cycles, SLC's cycle is generally 100,000 times, these values are not generally very significant, users would be able to write through the SSD itself as a measure of how many pieces of data standards. Through our most mainstream bars the diskless software systems - Rui Diskless XP 3.1, a load of work a hundred sets of diskless client server trace analysis, we found that in general, the average 24 hours to write the server the amount of data write-back disk around about 200G; we assume that the partial extreme cases the total amount of data written to 400G, and we found in the system then writes the data generated an average of 1.4 magnification, we also consider the extreme case enlarge the magnification of 3, this calculation down, an INTEL MLC 160G SSD can support the writing time of 4 years or so, do consider such a large amount of amplification, INTEL SSD are completely out of the cafe life of diskless server hardware!
e, SSD will bring in what diskless applications to enhance
If a server with 100 workstations, each workstation to the server per second to send a 64KB file in the hard disk read or write, then the server's hard drive needs to respond to 100 times per second per machine if sent the information 10 times 64kb read, the server will respond to the hard disk per second, 1,000 times per second to read about 64M, 15,000 rpm SAS However, 64KB data block random read test, the hard disk can only respond to about 160 times the average response time, 6.0 milliseconds (see Figure A), SSD can respond to 3000 times, average response time, 0.4 milliseconds (see Figure B), which is what a staggering figure.
Based on the above results can be seen that the speed of workstations to solve first to solve the hard disk I / O processing speed, or no matter how upgraded CPU, expanded memory, change the motherboard, there will not be a good role, the current In order to improve the lot of the server I / O processing speed to do the disk array, while the SSD solid state hard to achieve the performance, the most extreme cases, need to use the highest level of 10-20 to do the mechanical hard disk array, of course, is not in the diskless server could do with so much disk array, also do not need such high-performance, SSD diskless on the application, eliminating the need for multiple hard drives to do to solve the array of I / O bottlenecks!
the above analysis can not determine the SSD using a diskless server applications, the client-server with the machine to the Internet cafe to bring a significant amount of improvement, we have the actual test, SAS hard drives with the mainstream to do comparison, in the end to enhance the customer what we see with the practical application of Internet cafes, while a number of projects to do the test, test performance, the SAS drives are superior to the mainstream!
1, power test
front when the analysis of the characteristics of SSD to SSD talked about a diskless server, 1 hour power consumption is 0.07Kwh, SSD year diskless server power consumption is 613.2Kwh, that the 613 kWh, and 1 hour traditional diskless server disk power consumption is 0.13Kwh, the power consumption of the year 1138.8Kwh, ie 1138.8 kWh, then the SSD server diskless servers than the traditional one year, can save 525 kWh of electricity,
new balance 574!
2, the client start-up time test
start up when the diskless server to read and write is the most stressful, a single start, or the client all started when a small load pressure, the test speed of the virtual disk is no sense,
Species husband - Qzone log! we actually tested a size of 80 Internet cafes, in the Startup Items automatically restart a batch load, so that the client unlimited reboot the server at this time the pressure is very large, but the SSD is fully capable, start time with a single case of no load difference,
new balance running shoes! see the figure below:
traditional server hard disk configuration that in this case is how to show it? see below
can be seen on the traditional hard drive configuration the load is already high, Just look at the Internet cafe with a drive high-load conditions, the traditional hard disk configuration server, the client start-up time gradually increased, we can imagine, to the loss of customers is great, and SSD applications on the diskless, but to ensure that the interests of customers, in the case of high load, full load with a single start-up time did not differ!
the same time to test the project also includes the guest virtual disk random read parameters, network boot into the speed of the game, the game also boot into single speed, these, not the detailed data of the patients, with the most mainstream of the SAS server configuration contrast cases, are substantially improved!
diskless disk write in the early disks are a few pieces of machinery to do RAID 0, to improve the number of disk speed and IOPS, in the actual test, the diskless small data packets are randomly frequent operation, and the traditional mechanical disk RAID 0 in the random write test, write speed but decreased a lot, see below
ST300GSAS 15K7 single disk
ST300GSAS 15K7 3 disk RAID 0
not enhance the anti-
to drop the 81%
with mechanical disk IOPS numbers, and processing data for a long time (disk seek decision) bottlenecks, which is why there is sub-post write-back disk model, the more an advantage, then we see next SSD random write test, can not replace the traditional mechanical disk sub-task of writing and RAID 0.
IntelSSDMLC160G single disk
IntelSSDMLC160G 3 disk RAID 0
improved overall 230%
diskless seen the application of a single SSD, there is no concept of sub-wrote, the real breakthrough, this challenge is to get the real solution can be followed by a new problem has emerged, in theory, to do a single Intel SSD 80G 160G write disk, the performance in terms of performance, with 150 clients, all without any problems! In today's high-speed development of the Internet now, online video, games, leaving the growing network of more colorful, in the excellent state of broadband Internet cafes, clients continue to download, watch online videos, watch movies online using P2P software, a single disk to write the case of small capacity, is Kangbu Zhu pressure, but the problem did not bother us , how we look at the following address!
number of Internet cafes, we return to the country to write the conclusion by analyzing the data, IE, Youku, potatoes and other video sites accounted for a write-back write back more than 50% of capacity, PPLIVE, PPS, Thunder look at other online Cinema accounts for more than 30%, these two write-back capacity accounted for 80%, less than 20% of the remaining capacity is the write-back write-back games, we all know that online games are required write-back speed, and Online video is not required to write-back rate, SSD valuable space let these online videos to occupy!
As can find the problem, the problem is not so difficult to solve! sharp writing from the back and the IMG package file is associated, there are several IMG packages have several write-back file.
Next, we establish the sharp end from the service about a 10G TEMP image packages, IE on the client to the temporary folder, video software into the disk cache directory, note: Do not put the client in this virtual memory disk, because the game will be a lot more open when the speed of virtual memory used.
Then we use the sharp
written offer from third-party software tool to divert the sharp writing from the back of each image file to separate write-back to TEMP file to develop the package to the SATA (figure M disc) and write-back disk in the system and game disk write-back into the SSD (figure L plate) write-back disk. Online video cache files are basically seen edge existence, before the data stored in the hard disk has been used, stored to the hard disk after almost not using SATA drives to call it can competently complete!
L disk is the Intel SSD drive, M is an ordinary SATA disk, the client starts, sharp writing from disk streaming tool automatically diverted to the write-back data set specified disk
Finally, we watched on a 30-minute online Youku amount of data after the write-back, write-back system, only 64M, TEMP write-back has reached more than 500 M, the above method can write data back to SSD control of 50% in the previous, we recommend using two Intel SSD 80G RAID0 do to change the write-back according to the size of demand.
six, the future market outlook
increasing application of computer technology in today's diverse and unorganized, centralized management becomes more important, the concept of cloud computing is the result, in large part we can say that diskless software actually works in the enterprise established within a data set on storage and software distribution to private cloud, now more and more areas, including education, hotels, government, military, important business data, are in such a large number of systems to deploy their networks obtained data security and ease of management, diskless technology will certainly have a wider outlook. The software runs on a diskless hardware platform - the diskless server must also be made more demands for more, SSD solid state drive in the full application of the diskless server to the diskless server to bring the following advantages, performance, more powerful, cheaper, lower power consumption, higher reliability, compared with the conventional diskless servers can completely replace the existing server configuration framework, and become the mainstream!
todsgroup2hPEtod 2010picaudcarole Blog Archive l
Linear Mode
