Make sure that the problem is caused by servers Exhaust CEO and founder Jesse Rothstein said that many performance issues that seem to come from the server level are often caused by problems at the network, database and storage systems. Rothstein said: "In the end, you need to consider and analyze these issues in general, but it is actually tricky. "DarkKlimke, Global Product Marketing Manager at Fluke Networks, said: "If there is a delay, we need to see if this delay first appears on the server or at the network or client level. Tracking and analyzing symptoms takes a lot of time. & rdquo; IT departments often connect multiple servers to provide resources (such as applications) to end users. Klimke said: "You may have a web-based front-end server that talks to the application server, and the application server talks to the database server, all of which are connected through some kind of network. So the first step in troubleshooting is to make sure the problem really comes from a server, not from the network. ” Get the cumbersome application It is entirely possible that an application causes the server to respond slowly or simply shut down. Klimke said: "The application may be too sloppy & rsquo; (chatty) - that is, such applications are written without using large frames containing large amounts of data, sending fewer frames, but using The smallest frame, so the end of the day has to send a large amount of data back and forth frequently, in order to complete the task that a larger frame can accomplish; in contrast, the use of large frames is much faster and much more efficient. ” Clean up dirty networks Rothstein said that dirty networks are networks where packets are lost, reordered, or fragmented, which can drag down server performance, causing server performance degradation and server load. He explained: "Most people don't realize that network protocol stacks like the TCP stack have both fast and slow paths. The fast path is efficient, optimized for the packets it is to transmit, and the next packet can be transmitted in order. But if you transmit packets out of order, or inadvertently transfer packets from Layer 4 down to Layer 3, the server and operating system will consume a lot of resources to reassemble those IP fragments, or reorder or Reassemble the TCP fragment. & rdquo; Virtualization may complicate this issue. Rothstein said: "Your hypervisor is a scheduler …… share processor time across all running virtual machines and guest operating systems; if you switch from one virtual machine to another, it will inevitably appear delay. If the configuration is wrong, the network packet will be delayed, so that it is discarded by the underlying network stack; once you reach the tipping point, performance degradation is not only obvious, but also has catastrophic consequences. & rdquo; Using monitoring tools Rothstein said that no matter whether it is an automated third-party tool that performs SNMP polling to record server health indicators, or a relatively simple free open source tool, you should make full use of monitoring tools. Vollering said that if you are using a virtualized server, monitoring tools are especially important. He said that VMware, Microsoft and Citrix, the three major virtualization technology providers, provide excellent monitoring tools. He said: "These tools can tell you what is putting pressure on your environment, whether it's processor usage, memory usage or disk usage. " Vollering said that even if those tools are available and deployed, many companies often don't know if they can use them or don't know how to use them. In addition, these tools bring flood-like information and alerts. He said: "You may have to spend a whole day looking at the various charts. & rdquo; So, how can we make the most effective use of these tools? Vollering said: “Read the manual carefully and configure the tools for your environment. That way, if an alert occurs, you know that you need to explore the alert, not because the tool is still configured with default thresholds and is flooded. ” For example, a command from an application might be sent to the server, causing problems with the server. Klimke said: "Look at the actual data traffic entering the server, you can find out which link the packet entered, and what command was sent to the server before the server was down. This necessitates a view of the actual data traffic on the network connection, as well as the ability to interpret past commands and frames that were sent before the server was down. ” Checking the server's physical resources KellySather, CDW's internal solution architect, said that once you've identified a performance issue from the server level, you should double-check the physical resources of the problem server. He said: "The best case is as simple as adding another processor or adding more memory. "Dan Klimke, Global Product Marketing Manager at Fluke Networks, added: "It's important to look at whether your processor utilization is too high for many processes. We found that many problems were caused entirely by the disk being full. ” Make sure the server is up to work Sather says that because your hardware is compatible with newly installed or upgraded applications, it doesn't mean the application will run smoothly. He said: "The resources in your server need to be commensurate with the load it is going to handle. For example, someone moved to Exchange 2010, so they moved away from the old version of Exchange running smoothly on a dual-processor dual-core system with 24GB of memory. But because Exchange 2010 is a more resource-intensive system, they discovered the system in the old hardware afterwards. Can't run on. & rdquo; Correcting and adjusting the load The load from multiple virtual servers can far exceed the processing power of the physical server. Erwin Vollering, director of virtualization services at consulting firm GlassHouse Technologies, said: "You need to know your limitations." If you take a snapshot while processing monthly financial consolidation numbers, your storage system will not be able to handle all of those transactions at the same time. "The real-time migration of virtual machine CDW's internal solution architect KellySather said that as the workload increases and the number of virtual machines soars, you may find it necessary to migrate a problem-causing virtual machine to another host, or “Normal" The virtual machine is offloaded to another host to make room for virtual machines that need resources. VMware's vMotion and Microsoft's Hyper-V (and other technologies) allow you to migrate virtual machines to other hosts in real time, without the need to connect to end users.