若何处理无盘的回写压力(国外英文资料).doc

如何解决无盘的回写压力(国外英文资料) On the 15th of September, 2010, the article was posted on the article So from the article to now, we have made a lot of useful data, now lets conclude the what conclusions based on these data we can: In the week of testing, all the client back to write the amount is about 460 g a day on average, through the billing software, according to computer time is 759 hours a day, so at this time we can get: Each client to write data to the servers average speed per second (460 * 1024 * 1024 g)/h (759 * 3600) = 177 KB/S, but because is tested under the condition of no cache, in order to avoid to write speed breaking, cause the card machine problem, so we need to write back plate of the reserved three times performance allowance, which is 177 * 3 = 531 KB/S. Now lets figure out what the load capabilities of the next write disk might look like. The limit of a ssas disc is approximately: 11 * 1024/177 = 64, with three times the performance margin, the machine 64/3 = 21, and the three-piece disk is about 63. The speed limit of a raptors hard drive is approximately: 9 * 1024/177 = 52, with a three-fold performance margin, 552/3 = 17, and a three-piece disc with 54. The limit of a SATA hard drive is approximately: 6 * 1024/177 = 35, 3 times performance margin, 35/3 = 11, and 35 sets of disk drive. SAS disk, for example, the client and the server didnt cache, 1 piece of plate machine only 21 sets, so if 100 machines you need 5 piece of hard disk, 200 machines need 10 write back plate, when the machine sets of 500 units, you need 100 piece of write back plate to fix, this is too exaggerated, averaging $1 server to put 4 write back plate, according to the load capacity, we will need 25 server to process... Apart from investment issues, and even management is also very troublesome thing, that for the hard disk will adopt what kind of technical solution to solve the problem of diskless back to write? The result is a local write-back cache! First to get to know the l