标题: 求助：FASTt900 的带宽问题。 [打印本页] 作者: brian821003 时间: 2006-03-18 18:12 标题: 求助：FASTt900 的带宽问题。 请问各位高人。。FASTt900 带宽的理论值和实际最大值能到多少啊？
1. 如果是满配8个MINIHUB，感觉理论值应该是8G带宽，就算上损耗，传输速度最少也要有800MB/s吧？不知道这么说对不对。
2. 现在我在MDC上用 Sanyge 测试，速度只有160MB/s~180MB/s，而在其他服务器上测试也只有160MB/s左右。这正常么？
3. 在另外一个屋子里的机器同样使用 Sanyge 测试，速度居然只有80MB/s。这又是为什么？是因为走线的原因么？
4. 所有的光卡都是 emulex 9002 或者 982 的。。会不会是光卡的原因？
希望各位高人能够解答。。谢谢了。。作者: 宇风 时间: 2006-03-18 22:02
800MB/s 是指从ＣＡＣＨＥ中读取的速度，块大小是2048K的话，速度170MB/S是正常值．作者: brian821003 时间: 2006-03-20 09:51
呵呵。。那按道理说是不是所有的工作站都应该在170MB/s左右阿？现在问题是四台隔壁屋子的上载工作站速度只有80多阿。并且还有一个致命的问题，用这四台上载采集50Mbps的I桢数据流的时候，无论是几台并发都会抱错。
信息如下：The capture filter has prematurely aborted the capture because no more buffers were available. The data rate may be too high for the attempted operation. You must exit the application and restart your system.
大致意思就是说缓冲区不足。。。并且采集15Mbps的IBP码流的时候也会抱错。信息大致相同。。No further internal buffers are available to the capture filter
[ 本帖最后由 brian821003 于 2006-3-22 16:02 编辑 ]作者: wnzl 时间: 2006-03-23 09:52
关注ing，好像只有采集高码流才会有这样的问题作者: brian821003 时间: 2006-03-23 11:04
对。。只有采集50Mbps的I祯和15Mbps的IBP码流才会出问题。15M的出错几率很低很低，将近4个月只出过一次。就是50M码流，采集一小时左右必出错。。报错就是缓冲区不足。。头疼死了。。正在查手册。。祈祷吧。。作者: brian821003 时间: 2006-03-23 13:02
关于cache read ahead multiplier ，我查手册看了下，原文如下：
Read-ahead multiplier
This parameter affects the reading performance and an incorrect setting can have a large
negative impact. It controls how many additional sequential data blocks will be stored into
cache after a read request.
Obviously, if the workload is random, this value should be zero. Otherwise each read request
will unnecessarily pre-fetch additional data blocks. Since these data blocks will rarely be
needed, the performance is going to be negatively impacted.
For sequential workloads, a good value would be between 1 and 4, depending on the particular environment. When using such a setting, a read request causes pre-fetching of several sequential data blocks into the cache; this speeds up subsequent disk access. This leads to a fewer number of I/O transfers (between disk and cache) required to handle the same amount of data, which is good for performance in a sequential environment. A valuethat is too high can cause an overall performance decrease because the cache is filled with read ahead data that is never used.
Use the performance monitor to watch the cache hit rate for a logical drive to find a proper value.
我大致翻译了一下：
这个参数影响了读取的性能，不正确的设置会导致很大的负面影响。它控制了在发出读取请求后有多少个additional sequential data blocks要存进缓存。
显然，如果工作量是随机的，这个值应该为0。对于所有的读请求都不必预读additional data blocks.因为这些数据块很少需要，这些性能都不大必要。
对于有一定持续工作量的情况，对于特殊的环境1和4是比较好的值。当使用这种设置的时候，读取请求会把一些特殊的数据块读入缓存，能加速并发磁盘通道。这将导致少量的I/O传输（磁盘和缓存之间）需要相同数量的数据。哪些设置对于持续工作环境有利，过高的值会引起性能的降低，因为缓存会被那些从来不用的数据填满。
使用性能监视察看逻辑驱动器的缓存的命中率，从而找到一个合适的值。
The choice of a segment size can have a major influence on performance in both IOPS and throughput. Large segment sizes increase the request rate (IOPS) by allowing multiple disk drives to respond to multiple requests. A small segment size increases the data transfer rate (MB/s) by allowing multiple disk drives to participate in one I/O request. Use a small segment size relative to the I/O size to increase sequential performance.
You can use the performance monitor (see “Performance monitor data” on page 85) to evaluate how a given segment size affects the workload. Use the following guidelines:
1、If the typical I/O size is larger than the segment size, increase the segment size in order to minimize the number of drives needed to satisfy an I/O request. This is especially true in a multi-user, database, or file system storage environment. Using a single drive for a single request leaves other drives available to simultaneously service other requests.
2、 If we are using the logical drive in a single-user, large I/O environment (such as for multimedia application storage), performance is optimized when a single I/O request can be serviced with a single data stripe (the segment size multiplied by the number of drives in the array that are used for I/O). In this case, multiple disks are used for the same request, but each disk is only accessed once.
3、 Normally, a small segment size is used for databases, normal sizes for a file server, and large segment sizes for multimedia applications.
4、 If we increase the segment size we gain more throughput
