朋友们 我一直在遇到与数据库利用率或网络利用率有关的问题. 我们最终的应用程序软件保持冻结或在峰值时变得非常慢营业时间.尽管我们发现了一些需要时间的查询,但我们也已经成功地对其进行了优化.现在让我头疼的是,每当应用程序运行缓慢时,网络利用率就会随着数据库利用率一起上升. 鉴于这种趋势,我开始认为可能我的数据库不是罪魁祸首但是可能存在一些网络问题,由于该问题,一旦用户数量增加,应用程序就会变慢. (我们使用的是1Gbps网络). 如何检查问题出在数据库还是网络上? 我想一直监视这两个问题并证明谁是罪魁祸首. 我正在使用SQL Server2005. 请帮忙. 感谢 Mayur Kashikar
Hi FriendsI have been facing problem related to either Database Utilization or Network Utilization.The Application Software at our end keeps freezing or becomes very slow at the peak business hours.Although we identified a few queries that take time and we have been successful in optimizing them too. Now what's itching my head is, that whenever the application is slow the Network Utilization also goes up along with the Database Utilization.Looking at this trend I have started thinking that may be my Database is not a culprit but there can be some Network issue due to which the application slows down as soon as the number of users increase. (we are using 1Gbps network).How can I check if the issue is with the Database or the Network???I want to monitor these two consistently and prove who is a culprit.I am using SQL Server 2005.Kindly help.ThanksMayur Kashikar
推荐答案从网络角度来看,您可能最容易做的就是在此高峰时间内进行跟踪,并使用高级用户专家来查看是否有任何一个用户正在垄断带宽.可以在 www.codeplex/nmexperts 上找到顶级用户". 进行跟踪时,请请注意,在1gig网络上,鉴于大量网络流量,捕获机上的负载可能是个问题.我建议使用不带过滤器的NMCap并监视命令行,以确保未决计数在15分钟内不会继续增长.如果需要考虑服务器上的负载,则可以选择在交换机的镜像端口上进行捕获.在这种情况下,请记住打开混杂模式以捕获所有服务器流量.另外,根据流量,您可能会选择链接捕获或循环捕获.无论哪种情况,我都将使用200M-500M的大缓冲区. 一旦确定是否有客户机垄断了连接,就可以进一步深入研究跟踪以查看它们的作用.如果这是SQL流量,则可以在"TDS"过滤条件下进行过滤.看看流量.然后在对话树中找到客户端,以缩小每个客户端的工作范围. Paul Perhaps the easiest thing you can do from a network perspective is to take a trace during this peak time, and use the Top User expert to see if there's any one user that is monopolizing bandwidth. Top Users is available on www.codeplex/nmexperts.When you take the trace, be aware that on a 1gig network, the load on the capturing machine could be a problem given a large amount of network traffic. I would recommend using NMCap with not filter and monitoring the command line to make sure the pending count doens't continue to grow over a 15 minute period. If load on the server is a concern, you might opt to capture on a mirror port on your switch instead. In that case remember to turn on promiscuous mode to capture all the servers traffic. Also, depending on the traffic you might opt for a chained capture or a circular capture. In either case I would use a large buffer size, 200M-500M.Once you determine if there are clients monoplizing the connects, you could dive further into the trace to see what they are doing. If this is SQL traffic, you could filter on "TDS" to see just that traffic. Then find the clients in the conversation tree to narrow down what each is doing.Paul
更多推荐
数据库利用率与网络利用率
发布评论