Weak current College】 【broadband router performance evaluation of common errors.

Broadband and narrowband, the most important is a "fast", so we buy broadband router, is very concerned about its performance, popular, "fast". So many people made all kinds of evidence to prove their own broadband router performance is very good, very quickly. But a lot of evidence tend to be one-sided, and so you buy back found performance when huizhiwanyi insufficient. This article from a professional perspective, the analysis of several members often refer to it easily misleading evaluation of "evidence" that tell you how to objectively such evidence, and how to correctly judge the performance of a broadband router.

Evidence: the processor clock speed

Remark: the processor clock speed 100M, strong performance. "The processor clock speed performance ratio 100M 133M, much better. "" A communication professional RISCCPU. "
Myth: first, the processor is just a part of the router, it is so special about it is the core of the device. Processor use poorly router performance is good, but in turn processor ready router performance are not necessarily good. Like cooking, no costly properly fresh pork absolutely make a good's red-braised pork, but costly properly fresh pork, soy sauce is not good, not good cook, so make a good's red-braised pork. Secondly, the processor clock speed is only one processor, you want to know the performance metrics of performance must also know the bus width (16 bits or 32 bits), Cache size and structure, internal bus structure, is a single-CPU or multiple CPUs, distributed processing, operation mode and so on, these will have a significant impact on performance is anything but minor than 1 GHz. The so-called "adopts communications professional RISCCPU" is equivalent to the said nothing, just using terminology makes advanced, can't amount out, because almost all routers is communication professional RISCCPU, key depends on the CPU what exactly it is used, how the internal structure of the kernel.
Advanced users: generally speaking, 100M or less is low frequency, 100M ~ 200M secondary, higher clock speed 200M above belong to. In addition to ask what the kernel, the processor is 80186, ARM7, ARM9, MIPS or Cache capacity IntelXscale?? is a single-CPU or multiple CPUs distributed processing? note don't ask don't tell the truth of trafficking. 80186, ARM7 core processor is the first generation broadband router in a typical configuration, the performance is low, mainstream manufacturers were not used. MIPS kernel processor ARM9, is now mainstream. IntelXscale schema is an advanced network processor for high-end products. Cache capacity 8K or following falls, common, 32K 16K or above are great. General processor are single-CPU, a multi-CPU distributed processing is an advanced, high-performance processor. If you are a professional or like an inquiring mind, and you can get to the bottom of the ARM9 is ordinary 920T/922T/940T or enhanced 926E/946E/966E, MIPS is 2 k, 3K or 4K, 5K, different model performance and struct differences are larger. You can put the processor model entered into a search on Google, and then to the chip manufacturer's Web site, take a good look out, by the way, also look at other people's assessment of the chip (commentators, purpose level, cannot be taken with a grain of salt).


Evidence two: memory capacity


Remark: "I have this memory 16M, look at that before 8M, performance was much better than it. "
Myth: that there is a problem, two traps. Problem is: memory also is only a part of the router, it is an important part. And processors, memory use small router performance is good, but in turn memory used large router performance are not necessarily good. Memory is used for storing calculations all of the data in the process, using memory like layout of the room. Smart Shanghai women are good at "bellamya purificata shell do Dojo", useless things must not be placed in the room, everything is out of reasonable room in every corner, from floor to ceiling, every inch of have been fully utilized, some few 20 10 square meters of a House to put down the whole family all furniture and appliances, and even space on one or two bonsai embellished it. But if you change a lazy Bachelor, you give him a two bedroom flat and the same amount of furniture and electrical appliances try to ensure that he does not, but a few days, the room is full of junk, not based on the place. So according to the size of the memory used to absolutely evaluation performance of the router is not reasonable and scientific use of memory are equally important, high-level software design well planning and use of memory, lower or even no design capabilities, direct use of the processor chip manufacturers to provide the unoptimized reference software, memory will not be able to effectively plan and use. Two traps are: memory units, and memory bus. Memory can be used to do Byte (bytes), you can also use Bit (bit) units, a tone, capacity difference 8 times (1Byte = 8Bit), which may be of uppercase, lowercase B represents b Byte represents the Bit, but some people with ulterior motives and play tricks in here, like the above statement case no standard units to mislead, knowingly put b label do B the bait. Memory bus width is narrow, have to keep in memory or remove the data, if 4Byte is 16-bit memory, it is necessary to operate two, 32-bit memory you only need to do once, twice the efficiency difference. So even if the same amount of memory is not the same as if the bus width to router performance impact is significant.
Advanced users: generally speaking, 1M ~ 4MByte belong to small, medium, 16MByte 8MByte belong to or above belong to larger. Additionally if someone just tell you what is M, you have to ask is whether Bit, Byte is 16-bit or 32-bit. Of course you are looking for trustworthy object to ask, please don't be deceived. If you are a professionalPerson or the like, the memory model began to enter into a search on Google, and then to the chip manufacturer's Web site look out is never wrong, unless the criminals used grinding chips at the chip screen printing on counterfeiting. So select the honest code of honor of the manufacturers product is important.


Evidence three: Flash capacity
Argument: "my Flash capacity, performance is better than it. "
Myth: there is no reason. Flash is used for storing operating system and applications, and its size depends on the operating system, application writing efficiency and user interface of fancy degree. If you choose to use efficient real-time operating system (such as VxWorks, Ecos, Nucleus, etc.), designers and a deep understanding of the operating system, be reasonable cropping, designers to write efficient applications is also high, you can use the small capacity of Flash. If you choose to use inefficient operating system (e.g. Linux, etc.), designers do not understand the operating system, application writing efficiency low, or simply copying the chip manufacturers to provide the unoptimized reference software, you can only use the bulk of Flash. From this perspective, it is a Flash with as small software level higher, more reliable product. Of course, product functionality and user interface fancy (such as many high-definition picture) on the other hand uses Flash.
Advanced users: ordinary users never have to know Flash is great, as long as the watch function is not needed, so take a look at the user interface does not look good, doesn't fit your taste (if you have preference) is sufficient. If you are a professional or like an inquiring mind, you can try to ask the manufacturers they used what operating system, the software is written or copied, but basically you will touch a rebuff, but manufacturers are confidential. Evidence four: Throughput
Argument: "the strong performance, Throughput up to 97M. "" 64Byte packet Throughput to achieve wire-speed. "
Myth: the router works very much like the postal business. The post office parcel post process is that we get to post articles and the sender, recipient, and items of information to the post office, post office in different dimensions of the container and the items wrapped in the box on the format of the notice, to inspect and correct after posting to your destination, your parcel is the opposite process. Routers send and receive only basic, is data. Throughput (throughput) represents a router per second can handle the amount of data, corresponding to the post office unit time of package-handling capacity, is a Visual reflection of router performance. It is important to note that the data behind something is up.
First of all should indicate the so-called router Throughput, must have the Throughput, data flow LAN-to-WAN out or into the LAN router processing is necessary to the performance of the router. Instead, the representative of LAN-to-LAN is the router's internal small switch performance, normal-speed should have been, there is no sense. This is like the post office unit time of package-handling capacity, is shipped and transported into the post office parcel volume, rather than from the post office is a room to move to another room package rate. But still some data as router LAN-to-LAN Throughput in publicity, even marked on the product packaging, and of course he wouldn't write this is LAN-to-LAN, we must be careful.
In addition, the router Throughput, generally should be opened in the NAT, firewall turned off the test data. This is because the NAT is a broadband router the most fundamental and core functionality, do not open the NAT does not make the broadband router, and although the different product design the NAT function is the same, but the software design direct impact to the NAT router performance, efficiency, so NAT open Throughput is meaningful. As for firewall, should be counted broadband router fringe functionality, advanced capabilities, there are many products firewall rules are complex and can filter a lot of things, there are very few product rules and simple. Rules for multiple, complex, CPU use to filter data for a long time, the rules for small, simple, CPU use to filter data in a short time, the effect on Throughput test data or very large. In the interest of fairness, in test router Throughput, especially when comparing different product performance, turn off the firewall. This is like the anthrax mail attacks in the United States occur, the post office on the postal check strictly a lot, this will have on the post office units of time for postal processing capacity to produce more impact, but other countries do not have such a complex examination procedures, if we assert that the United States accordingly that inefficient post office, it is unfair. As for firewall evaluation, a general comparison of the functionality inside instead of the performance comparison. Of course, if it is not a comparison of different products, only on the performance of a particular product itself, if both have the firewall off and turn on the firewall (which needs to be marked Enable filtering rules) in both cases the Throughput data, we can understand this product more thoroughly.
Common Throughput test method has two, one is a test, Smartbits is Chariot test. Smartbits test is to use the world's most authoritative network equipment testing instrument Smartbits2000 or 6000, with its accompanying professional test software SmartApplication, continuous UDP packets to detect on different size of packets to a router per second can handle the number of packages. Router in the address of the packet, the main time is spent dealing with end of baotou, packages, depending on the size of packets to a router per second can handle the number of differences of the package. This is like the post office handling packages, the main time is spent in the processing of the sender, recipient, and article information and inspectionOn the package to bit weaker influence on processing speed. If 128Byte packages can process per second and 10000 cannot do 64Byte packet per second 20000, but only slightly more than 10000 a little bit like 10100. This gave some dishonest manufacturers to advantage. As his router in dealing with the biggest 1518Byte package 000 per second, based on theoretical calculations processing 1518Byte package 100M wire-speed limit value is converted, so 8127 out of Throughput is 100M * 8000/8127 = 98.44M, hence he is very proud to announce that my router Throughput up to 98.44M, everyone knows that the router in the smallest package 64Byte is 11000 per second, based on theoretical calculations processing 64Byte package 100M wire-speed limit value is 148810, converted from Throughput only 100M * 11000/148810 = 7.39M, both differ by more than 13 times. It all depends on how we think throughput is how many per second per second packet or the number of megabytes of data, like post office every day is processing many parcels or how many kilograms of cargo. From the post office capacity point of view, should be dealt with from the daily number of packets to be judged more reasonable, since this data is relatively stable, small parcel weight impact. But how many kilograms of goods per day, subject to the weight of the parcel is fundamental, such as mailing of all books, calculated daily cargo handling large number of kilos, if all cotton is very small. Reality is both a book and a wide variety of cotton and other stuff, but when you assess what percentage of the allocation of the goods is only reasonable? no one made it clear that unless someone has a standard provides a variety of things, we all follow this standard. Since no standards, it has been used books to calculate, calculate with cotton, the rules of my what counted, not the rules on happy fish in troubled waters. In addition to current broadband router generally used by the hardware, even the best performance, but also insufficient to achieve wire-speed packet Throughput 64Byte, but some people claim that my product is so good, but take out Smartbits test data to prove. This type of product is the so-called NAT hardware acceleration technology, Throughput is nice, but at the price of all the data is not passed the CPU, not data filtering, open the portal, the firewall does not work, the router does not have any management functionality.
Chariot test is used for two servers, install the NETIQ Chariot of popular software company, a received router WAN port, a single LAN interface that received through statistics a predetermined length and format of the script file without error from one server to another server to calculate the router Throughput. This scholarship is big. First, server performance, if performance is poor, will be a test of the performance bottleneck, the router is not reflected, so dishonest manufacturers in testing their products with a high-performance servers, lest their product performance, and test other products with a low performance of the server, deliberately derogatory data. The second is to establish a connection, if the router software written in poor, increase in the number of connections, Throughput is plummeting, and good router is very stable, and will even increase with the number of connections have slightly increased. If the router is not enough, even the performance will be returned as well as the setting up of so many connection and the test fails. So bad routers only dares to use a connection to conclude that the test data, while good router does not fear the number of connections. The third is the Chariot test does not test that is testing Smartbits UDP "package" instead of "flow" of data, the data is encapsulated into large TCP packets and server operating system related, but are generally large package encapsulated as 1518Byte, so its test data will be used to test small packages 64Byte Smartbits data pretty many times, this can confuse people. Fourth test Throughput results Chriot is a valid data load, does not include TCP protocol loss, frame interval, answer and Chriot itself system loss, this part of the typical loss calculation based on the theory, i.e. about 6M even if you add a table to the switch at forward, is not a router, logging out of Throughput and can only be around, this is 94M theoretical limits, if someone claims that my product Chriot test Throughput97M, 98M, that there is no doubt is absurdly.
Advanced users: Throughput is the Throughput, General LAN-to-WAN should be opened in the NAT, firewall turned off the test data and test and Chariot Smartbits test in two ways, the results can be very different. Under the current common hardware capacity, Smartbits large package test data we are easy to achieve wire-speed, just like you to test students ' intellectual, 1 + 1 =?, and everyone can answer, than not. So you want to make an objective assessment, the title should be a difficult one, compare 64Byte small package test data, high stand convicted. Of course, you should be careful if there is a NAT hardware acceleration, so to turn it off and then test. As for the Chariot test, the performance of different products should be in the same test environment and methods, the same number of connections is only fair to compare, but in the actual network applications, always more connections and almost impossible to be a single connection, in order to find bad routers, Chariot test is best carried out in a number of connection, the General can choose 100 to connect basically can see everyone's differences. In short, separate a Throughput data is meaningless, be sure to indicate the data is what method is only useful for measuring out, mutual ofThe comparison must be in the same test environment and compare is under way in a fair and meaningful. The best way is to look at Smartbits test NAT on small Pack LAN-to-WANThroughput 64Byte.


Evidence number five: belt machine
Argument: "this router machine quantity is 200. "" This router maximum allowed amount with machine to 253 computers.
Myth: it is not accurate. How busy every network, Internet cafes in all of us are obsessed Internet chat, games, and almost all the data through a router WAN, so heavy loads. But if you are an enterprise network, most people are busy engaging design, writing reports, make plans, at the same time only a small number of people use networks, but most of the data flow in the enterprise network, the router load is light. In a 200 PCs enterprise network performance adequate for the router, Internet cafes often may even bring 50 PCs. Estimate a network each PC average data traffic is not so precise. This is like cooking in the canteen staff, can only be estimated per capita have four two, this one pot four pounds is enough for about 10 people, it is likely that one day we do a surprisingly moving, appetite, a pot of rice with only 6 people, or what day of the flu all appetite is generally not good, then 5 people is enough to eat. Miners ' canteen mentors to estimate the average person to eat six two refectory, and the female model master may only dare to estimate the average person to eat one or two. Therefore, more objective statement should indicate the amount of this machine is for which type of network and number is an estimate based on a typical scenario, such as the scope of "Internet machine of 150 ~ 250 (typical value)", it is responsible. The second largest amount with machine is bluffing, it is not based on the performance of the router, which is the maximum you can DHCP assigned IP address number, 254 knock yourself out of a used is 253.
Advanced users: machine number is only an estimate and experience values must be combined with the actual status of the network view, accurate performance depends on the test data, test the NAT open Smartbits 64Byte small package LAN-to-WANThroughput is pps (packets per second). But with the number of regular users is very intuitive, very well understood, we refer to this data is important to note that the above mentioned several mistakes. In addition this data if it was from the major manufacturers, reputable manufacturers, general statement comparing the rigorous, high reliability. If you are coming from small manufacturers, kitchen manufacturers, general statement ambiguous, the credibility of the poor.


Evidence number six: WAN port
Argument: "this router is a dual WAN port, performance is a single WAN port to two times. The "single WAN port with machines of 100 units, Dual WAN port with machine quantity 200. "
Myth: the confusion of concepts. A router-based hardware and software, its processing power or performance is identified, does the number of mouth with WAN and changes in the larger change. In one case: router itself handling capacity relative to the WAN port bandwidth there are extra processing power, such as routers, WAN Hau 40M bandwidth per line 10M, limited bandwidth, single WAN port router only 10M throughput, Dual WAN port router you have 20M throughput. From this perspective seems to improve network performance has doubled, but it's just your network configuration is reasonable, the router performance has always been, no changed 40M. Conversely, if the router itself handle only 5M, whether a single WAN port or Dual WAN ports are only possible with 5M throughput. For example, a factory can do ten thousand pieces a day, but only a cargo truck, daily delivery of five thousand products only when adding a cargo truck is very effective. But if the daily yield only 3000 pieces, also equipped with two cargo trucks, no role, increasing costs.
Advanced users: now there are many different brands of some WAN port router in sales, but performance varies greatly. A router doing multiple WAN port, it is first necessary to establish in the router itself strong enough performance to premise, relative to the bandwidth router processing power has the surplus, if their processing power is limited, many WAN port is purely a display. We select multiple WAN port router, be sure to carefully review their performance, it is available for your reference: If the router is ARM7 core or equivalent performance processor, clock speed is less than 100M, basically can determine its performance is not enough to do multiple WAN port.