| Back to logs list
22128020 2009 年 06 月 16 日 11:23 Reading (loading. ..) Comments (0) Category: Computer Related
〓 〓 〓 〓 release all the RAM disk on the network and how to set 〓 〓 〓 〓 〓
5.6 0359 5.8 on a lot of technology still continues. Cache 0359 is still very effective, believe the use of moving slow, you can greatly reduce the server stress
5.8 version of the friends not to try it with 5.6
0359 server to read the cache:
2G: None.
4G: 1024M dynamic (if not the client C disk optimization, more content can be dynamically created 1536M, can not exceed this value, that is 1.5G. If you have no effect, the server disk is still stubbornly high, then do not
, the go to optimize your C drive), page size: 64k (with the array area the same size cassette, onboard sata array of general 128k, scsi / sas is 64k), the cache scope: the size of your C drive, This range is less than equal to the circumference C
disk size (such as C drive is 5230MB, then the cache scope is 5G).
8G or more: single open image package to build 3G mirror package, open the cache after uploading your system as a static, 3072M.
multiple image packages, you can only open dynamic, each packet 1024 or 1536, pay attention not to meet or exceed the value of the cache memory of the half.
note, the game disk do not open any time the cache server, so as not to affect the game update the platform, in fact, read the game disc pressure is not great, do not look the game more.
client write cache:
single physical disk expansion method (for dynamic cache):
512M: Read does not open, write 64M. Page: Reading 0k, write 8k.
1G: reading 64, writing 128 or 96 to see customer needs. Specific conditions. Page: Reading 64k, write 8k.
2G: Read 128 Write 256. Page: Reading 64k, write 8k. In fact the client memory to 2G, the server disk has almost no traffic.
3G or above: Read 512 Write 512. Page: Reading 64k, write 8k.
physical disk mirror package + game:
512M: image package: Read does not open, write 64M. Page: Reading 0k, write 8k. Game board: Do not open any cache.
1G: image package: read 64, write 128 or 96 to see customer needs. Specific conditions. Page: Reading 64k, write 8k. Game board: Do not open any cache.
2G: image package: read 96 write 192. Page: Reading 64k, write 8k. Game board: read 64 write 96. The main part of the cache to this ^ ^ QQ software. Page: Reading 64k, write 8k.
3G or above: image package: Read 256 Write 384. Page: Reading 64k, write 8k. Game board: Read 128 Write 256. The main part of the cache to this ^ ^ QQ software. Page: Reading 64k, write 8k.
multiple physical disk image package + open method above. As for the local cache, it will not conflict with other packages.
S disk swap cache:
16G or more server memory is limited to use, can further improve the smooth flow of clients, completely remove the server to write the array pressure.
16G: each client 64M.
32G: each client 128M (theoretical value, not practice).
page: Unified 64k, set to maximum.
load per server in accordance with the range of 150 clients, two SAS disk write-back of environmental settings.
by the program to field test (server 32G expansion method excluded). S swap disk cache on the server within a single load 120 Shihai no need to open (if the client has opened more than 96M write slow), the effect is not obvious.
In addition, open exchange of buffer S, the server cpu usage will be increased, so 16G or more servers, you must use 4-core or dual cpu, dual dual-core cpu which has better performance and relatively lower occupancy rate .
ps:
1, only for the speed, the client-side caching of pages read can be modified to 0k, will reach a very high speed, but in fact the client will lose the effect of the local read cache. So customers end gun shows, after re-adjustment of the nominal value shall return to this page 64k.
here to explain the creation of client-side read caching, the server will do the speed limit for each client in order to play the role of load balancing, each client speed 60MB / S. This rate has been very comfortable. Read the client cache is set to 0k, that you actually read off the client cache, then the server to the client to consider all the data must all read from the server side,
tory burch on sale, so the client will no longer be the speed limit, allowing each Read full speed machine, which will naturally be high speed.
need to explain, if the client memory is small, does not intend to open the local read cache, then be sure to set the local read cache page, 0k, or the server that the client has opened the local read cache, and thus limit the client speed.
2, cafes bad if the machine a good machine, you need to observe:
good machine if more than poor little machine, then taken to a compromise, such as a large number of Internet cafes and a few clients 1G 2G client even a very small amount of 512 clients, the time to write 64 192, because less pressure to read the disk, so focus on write caching would be better. If only a few good machines to attract customers, the majority of 1G or 512 clients, then you can open read 0, write 128. At this point have to take care of the weak.
further questions please contact with customer service or test
following illustration shows what the effect after opening. Conditions for stand-alone test, the server 8G, 3X500 es reading,
tory burch boots, 2x73 sas write. Because of its role in the static cache is a client server C drive as a whole into memory, without using any third-party caching tools, so here only to make the test for the C drive to achieve the effect of the static buffer, to achieve the memory disc function.
published this chart is not how to prove what speed, a machine will look like. The first map, only the test C drive, a closer look, not the S plate. This is the hd tune. When the speed, net output is close to the limit card, disk reading of 0.
the first two maps,
tory burch reva, only the test C drive,
tory burch 2011, a closer look, not the S plate. This is the hd tach, 8m district. When the speed, net output is close to the limit card, disk reading of 0.
the first three plans, only to test the C drive, a closer look, not the S plate. This is the hd speed, 256k small page. Many people hate this, that it is optimal for all networks, in fact, the software is still in front of several test showed good results, but when the gun, net output is close to the limit card, disk reading of 0.
As today can only upload 3 attachments,
tory burch outlet, then the remaining mass after the needs of 0:00. Open and there are specific laws capture the cache. Note that the server and the client is not using any third-party tools to achieve the cache, so less easy network management technology for a variety of daily maintenance.
addition of limited because of this the next time when the display server memory 16G, S disk swap cache, write cache server is set up after the results.
posted instructions over and over again anyway, find someone better than others affixed to the top better, I tried a little bit 5,6 test out Internet cafes, also I tried a collapse, the price is not small, Oh, in a few maps, on the open method and 1.5T solution
this method is the open system disk is mirrored package, because only 3G size, then the static open 3072, the whole into memory, and this package in the server independent of the system tray, 80G inside, and that was not delay performance.
physical disk and then the first area is actually a little more than 3G, used to back up this package. Million accidentally redo server system, to get rid of the image package, as long as the physical plate made of this area re-image package on the line, tens of seconds to do, and the physical disk image packages from the system can double insurance.
this is the game disk expansion method, in fact, is what is inside the QQ leave room for, mere to the game as a read cache results that are not. Since the reading by MFU algorithm, so the game disk only the most commonly used QQ, hit rate is also this.
write cache is still the MRU algorithm, then slow enough to write this game disc QQ and some online games use. Write-back to play the game itself is small, and most or the page file to the C drive or write, so set enough.
the effect of this is the 1.5T, a mirror package, 3G, physical disk with three 500ES, for many games, I gave the customer to 1T, leaving the remaining space according to the customers themselves need to download.
3X400, SAS cost is too high, 4X146 also not much loaded games, speed is not faster than the 3 reading, and independent to take a server as a virtual disk mounted game performance much, more importantly,
tory burch flats, costs of electricity, and more accounted for a machine, also increased a little maintenance. General told, 600 dollar es low cost disk, performance, it is most at least better than the 10k scsi, support ncq, 32M cache, read cache and memory disk with the words, the performance was more than just a lot better test SAS. After all, 70% of the time pressure from the C drive.
customers now have scheduled 3X 1T es.2 with my hard drive. Do not know if he wants to pretend to be so great. We support the NTFS single zone 2T, linux single zone 32EB (1000T = 1E) data, there is no room who support bigger than us.
2 disk has enough of a write, it is fear on the last 3 months, even if someone intentionally you extract the bar, copy the file is not afraid to write disk resources are basically left him with, and this design is very good , not to mention opening the cache speed limit there is 60 trillion. Some say the low rate of 60M? I do not believe his kid to play full-duplex 60M is not enough.
a server that does not have much money, the game has more speed better, I usually bring 80-100's also so with:
Xeon5130/8G/3X es 500G read / 2X73G SAS write / Integrated Dual LAN shunt.
80 units within 3 500 3 250 write read enough, not to any independent game server, think about how much did this cost.
and other diskless difference is, win version of the technology unless you maintain a high bar, or would not dare to do supercache es on the disk, sata, after all, in the general performance under win2003, and 10k scsi for the same. Diskless linux other than temporary does not support 560G partition, not a good array, usually with two 500 es read, as large capacity, and speed on down, over 80 machines can not, or our 3 es500 1T R0 array is completely set, the performance or the best, combined with the cache, and do not block a cpu, that will leave customers with maximum performance. Some people say that does not affect the performance of single core, then in the past, if you meet both open QQ1 Warcraft 3 in the open on the top is not a movie over. This situation near the school too much.
Finally in the last few days, I figure, the last sold 16G server, the server I am on a 16G disk cache open S to see the plans and results, S disk is for the page file and the temporary folder using IE , and itself has been shared on the server side cache, S disk cache after the open, in fact, each machine on the server side to set aside a fixed space in their pages as a temporary folder caches files and use, so that 0, the server can achieve the basic write-back. Because you make each client machine memory is not likely to 3G, the cafe is not good, dried, 100 more than 100 dollars machine is not a small number, but you allow customers to buy 16G server memory he is willing, not more how much money.
Cheap newport cigarettes for you-(63)
Linear Mode
