Enterprise experience sharing with VMware vsphere for server virtualization

Enterprise experience sharing with VMware vsphere for server virtualization

For many small and midsize enterprises, virtualization has become an important means of saving IT spending, following the company's VMware Virtualization Product deployment case, to share with you in the use of vsphere products to build a virtualized data center, some small experience, You want to help with enterprise virtualization products and server and storage options.

First, the choice of virtualization products

When building a virtualized data center, virtualization product selection is critical, and we have always chosen VMware VSphere as a professional virtualization solution provider, but in the meantime, we have chosen other products, but there have been more or less problems, One attempt to use another virtualization product did not have any problems during the installation of the configuration, but a problem was found during the simulation of the production environment, when the user accessed the server's virtual machine from the LAN and the Internet, the speed of each connection was limited to 700KBPS/S, while the server's virtual machines, The speed between virtual machine to physical host is normal, can reach 7, 800 trillion. Later we follow the same environment, testing on other servers, found that some servers access the virtual machine through the network normal speed, some servers are limited to 700K more than 1M, and then communicate with customers, or replaced by VMware ESXi, the last network normal.

Servers using VMware vsphere virtualization are not only very stable, reliable, and secure, but VMware ESXi virtualization cores are few, can be installed on a 1GB USB flash drive, and for many years, in addition to product version upgrades and necessary patch updates, VMware ESXi virtualization hosts do not need to be restarted-VMware ESXi can run for years as long as the room is not power-off and not upgraded.

Ii. How to remove the doubts of small and medium-sized enterprises before virtualization deployment

For the SME users, before the implementation of virtualization, there is always a more or less concern, especially the existing business has been running smoothly for a long time, more mature information enterprises. Their biggest concern is whether the entire system is secure after virtualization is implemented. For example, the original unit has 100 applications, each application on 1 physical servers, the entire system has 100 physical servers, after virtualization, each application on 1 "virtual" servers, each host has 10 or more virtual machines, This allows 10 or fewer physical servers to be virtualized instead of the original 100 servers. The original server is broken, at most only affect the application; now if a server goes bad, it can affect 10 or more applications, which can put a strain on the information manager. Also, is the performance sufficient to migrate and merge multiple servers into one host?

In fact, the above concerns are nothing more than security and performance issues, the following Kai-tech one by one for everyone:

security issues: in the absence of virtualization, a server corresponding to an application, which seems safe, the actual risk is relatively large, a server is the corresponding application, from the quantitative point of view, spread the risk, but now this server uses "local hard disk", all of this server application is in the " Local hard disk "on, once the server crashes, the system can not enter, then, in the case of the server is not available, how to copy or migrate data from this" server "to other available host? This is not the same as our normal workstation, the computer is broken, the hard drive is removed to the other machine can be used. To know, any important server, is a number of hard disks and do raid, whether the simplest RAID1, or RAID5, RAID50, RAID10, are multiple hard disks, and simply the raid configuration of the hard drive to another server, not necessarily can be used, There is no guarantee that the data will not be lost (this requires higher technology).

and using virtualization technology, each server for multiple applications, compared to a server corresponding to an application, from a number of comparisons, it seems unsafe, but not only a simple digital contrast, from the use of virtualization, we should understand a principle: data and system (or application) separate. Simply put, all of the data is not stored in the "local hard disk", but is stored in a highly secure "shared storage", in the virtualized datacenter, the server is not local hard disk, or the configured hard disk is only used to install the virtualization system software (VMware ESXi) itself, and the virtual machine running on the virtualization layer, the data is stored on the shared storage, the server "local" does not save the virtual machine data, it is like in a unit, for each room to configure a air conditioning, or use the difference between central air.

In most enterprises before virtualization, most of the servers are configured with 1 CPUs, 2 hard drives to do RAID1 or 3 hard drives to do RAID5, single-supply, single-network (server two NICs only one piece), and in the virtualized project, the virtualization host server is mostly configured with four CPU, 6~10 or even more hard drives do RAID5 or RAID50, RAID10, four power supplies, 4 or more physical network card redundancy. Each server in virtualization has redundancy, redundant devices take over when a single network card, hard disk, power supply, or even CPU problems occur in the server, and in a virtualized project, shared storage is common, and virtual machines are stored in shared storage, even if a host is completely damaged. Virtual machines running on this host are started on other physical hosts, ensuring that the business system is not interrupted.

again performance problem: a single application host, mostly configured 2 hard disk to do RAID1, or 3 hard disk to do RAID5, so disk performance is relatively low. In a virtualized host, 6 or more drives are typically used, with RAID5, RAID50, or RAID10, with high disk performance. In addition, while virtualization runs multiple virtual machines on the same host, these VMS do not require a higher CPU and disk, memory utilization at the same time. Based on years of virtualization implementation experience, virtualization does not reduce the responsiveness of the original application to a slight increase.

Third, enterprise virtualization steps

In the enterprise implementation of virtualization, most of the first virtualization of the less important physical machines, migrating these physical machines to a virtual machine for a period of time (typically 1 weeks), to see whether the business application after virtualization, and to simulate some failures, to experiment with the problems that may occur later, and so on, after these tests are completed, And then virtualize the other physical machines. Some applications that are not suitable for virtualization will still run on the original physical host, such as servers for video-on-demand, important database servers, and so on.

Iv. how enterprises can utilize today's device architecture for virtualized environments

In the process of virtualization, most of enterprises will consider whether the current enterprise server, storage, switch and other basic equipment can be used, in fact, according to the server, storage performance and parameters of comprehensive consideration.

In the case of a newly purchased server for nearly one or two years, consider consolidating and expanding these servers to be used as virtualization hosts, and generally most of the servers that are not as standard can be expanded to a very high configuration. For example, the IBM 3850 X5 Server can scale up to 4 CPUs, 1TB of memory, dual power supplies, CPU, for example, IBM 3850 X5 factory standard 2 CPUs, this CPU can be 6 cores, 8 cores, if the enterprise has more than one IBM 3850 X5 server (such as 2 or more), The 2 CPUs can be placed on one of them, while the other can buy 4 8 cores of the CPU, the same, the memory can be centralized to one, the other is configured with multiple 8GB of memory, similarly, for other manufacturers of servers can also be handled, such as  rh2485 (1), Standard 2 CPU, up to 4 CPUs, 1.5TB of memory, 2 redundant power supply, can be multiple units to consolidate, and then the server upgrade.

In the process of virtualization implementation, if you use an existing server, it is recommended to add memory for the server, network card, followed by the configuration of redundant power, CPU, as for the hard disk, in the Enterprise virtualization project, the priority is to configure shared storage, followed by the addition of local hard disk.

In addition to the virtualization host, you can also change the original server storage server, for example, if a server configuration is low and does not have the value of upgrade, but with more local hard disk, you can set the hard disk to a server, Make this server a storage server by installing Openfiler (either 32-bit or 64-bit products) or Windows Server R2 or Windows Server 2012, providing iSCSI networked storage for virtualized environments over a gigabit network These stores can be used for data backup or extension.

Five, server performance and capacity planning

In the early stages of virtualization, there is a virtual machine capacity plan, which is the maximum number of virtual machines that can be placed on a physical server. In fact, this is a comprehensive problem, that is, to consider the host's CPU, memory, disk (capacity and performance), but also to consider the required resources of the running virtual machine, in actual use, the system always has at least 30% or more of the surplus capacity, it is not possible to make a host of resource utilization of more than 80% so close to 100%, Otherwise, once these values are reached, the overall system response will be slow.

When estimating the capacity of virtualization, in the case of CPU-only, the physical CPU and the virtual CPU can be 1:4~1:10 or even higher ratio planning, such as a physical host with 4 8 cores of the CPU, in memory, storage enough, according to 1:5 of the scale, you can virtual out 4x8x 5=160 a vcpus, assuming that each virtual machine requires 2 Vcpus, you can create 80 virtual machines. In a project that actually implements virtualization, the CPU requirements of most virtual machines are not very high, even if the virtual machine is assigned 4 or more CPUs, but in fact the virtual machine CPU utilization is only 10% or less, this time consumes less than 0.5 physical host CPU resources. 2, this is a virtual machine with an excess capacity allocation using Vcenter Operations Manager statistics, as can be seen from the list, most virtual machines have less than 10% CPU utilization and actually use less memory (although allocating 2GB of memory to most virtual machines, But the actual use is only between 256~576MB memory).

Figure 2 virtual machines with excess capacity using VCOS statistics

In a virtualized project, the memory footprint is the largest, the most demanding, and in practice, the administrator will find that the physical host memory will be close to 80% or even 90%, because on the same physical host, the number of planned virtual machines, And each virtual machine allocates a large amount of memory (always exceeding the memory that the virtual machine actually uses), causing the host to reduce the available memory by 3, which is the CPU and memory usage of a running VMware ESXi 5.1 host.

Figure 3 ESXi Summary in an enterprise operation

When you configure memory for a physical host, consider how many virtual machines will be running on that host, how much memory is required for those VMS, and typically, each virtual machine requires more memory in 1GB~4GB or more, and a portion of the memory is reserved for VMware ESXi. Typically, a host configured with 4 8 core CPUs typically needs to configure 96GB or higher memory, and in the case of a host configured with 2 6 core CPUs, it is usually necessary to configure 32~64GB memory.

Vi. statistics and calculation of existing server capacity

If you are migrating an existing physical server to a virtual machine, you can make a table that includes the current physical server's CPU model, number, CPU utilization, existing memory and memory utilization, number of existing hard disks, size, RAID, and usage, which are then calculated as Figure 41 is a unit of the existing server statistics (in actual case, the unit server about 100, Figure 4 and later Figure 5 is only a list of some of the server model and Resource usage statistics).

Figure 4 Utilization of existing server resources in a unit

Based on this, we calculate the actual CPU, memory, and disk space required for each server, calculated as:

Actual CPU resource = CPU frequency of the server XCPU number XCPU usage

Actual memory resource = the server memory x memory utilization

Actual hard disk space = hard disk capacity-remaining space

For example, the first server in the table requires 3.0ghzx2x15%=0.9ghz, memory is 2GBX30=0.6GB, and the hard disk is 73GB-30GB=43GB.

It is then calculated after Figure 4, which is actually shown in scenario 5.

Figure 5 Actual resource usage and last resource statistics per server (only partial servers listed)

After calculation, the project has used 91.1944GHZ CPU resources, CPU frequency 3.0HzCPU For example, you need 30 cores (load 100%), but to consider the overall project of the CPU load rate of 60%~75%, and management and other costs, You need at least 40 CPU cores, and if you configure 4 6-core servers, you need about 4 physical hosts. At least memory, now used 182GB, plus management and surplus, to 360GB computing, per server 96GB~128GB can.

If you do not buy a new server, and choose 4~8 High-profile server (for example, 6), the 100 servers using virtualization technology, the migration to 6 of them, the savings in electricity (400W per server, industrial power consumption of 1.1 yuan/degree) about 346,900.

If you want to use an existing server, you need to expand the memory for some of the servers that do the virtualization host. With an existing server, if you do not expand the CPU of an existing server, it is advisable to extend the memory to 64GB on 2 CPU hosts.

As shown in Figure 5, the 6.9T capacity is used to plan 10TB and even more storage for the entire virtualized system. In the case of backing up the original server data, it is possible to centralize the 300GB, 146GB hard disk to the virtualized host and use it uniformly. In the case of a virtualized host using 6 physical servers, each server needs 1.5TB~3TB space. When using RAID5, 1.5TB usable capacity is available with 6 blocks of 300GB, and 1.8TB usable capacity is available with 8 300GB RAID50. 3TB capacity is available with 12 300GB hard drives and RAID50.

Vii. Server selection when server virtualization

In the implementation of virtualization, if the existing server can meet the requirements of virtualization requirements, you can use the existing server, if the existing server does not fully meet the requirements, you can partially adopt the existing server, and then purchase a new server.

If the purchase of new servers, the choice of products more, if the unit room in the cabinet storage, priority procurement Rack Server, server procurement principles are:

(1) If the 2U server can meet the demand, then use 2U server, usually 2U server maximum support, 2 CPUs, standard 1 CPU, at this time, it is necessary to configure 2 CPUs.

If the 2U server does not meet the demand, then the server with 4U, typically, 4U server maximum support 4 CPUs and standard 2 CPUs, when the acquisition of servers, the server configuration 4 CPU is appropriate, if the number of servers do not limit, Purchasing twice times the 2U server saves more money than purchasing 4U servers, and most of the performance can meet demand.

(2) CPU: in the selection of CPUs, the choice of 6-core or 8-core Intel series CPU is appropriate, 10 cores or more cores of the CPU is more expensive, not recommended choice, of course, units of CPU performance, space requirements, except.

(3) Memory: When configuring the server, it is possible to configure large memory for the server. In a virtualized project, memory is more important than the CPU. In general, 2 6-core 2U servers are configured with 64GB memory, 4 6-core or 8-core 4U servers configured for 128GB or more memory.

(4) network card: in the selection of servers, but also to consider the number of server network cards, at least to configure the server 2 interface Gigabit network card, recommended 4-port Gigabit network card.

(5) power supply: Nearly two power supplies may be configured. In general, the 2U server selects 2 450W power supply can meet the demand, 4U server select 2 750W power supply can meet the demand.

(6) hard disk: If the virtual machine is saved in the local storage of the server, instead of the network storage, configure 6 hard disks for the server to do RAID5, or 8 hard disks to do RAID50 appropriate. Because the server hard disk slot is limited, it is not possible to choose too small hard disk, the current cost-effective 600GB SAS hard disk, 2.5-inch SAS hard disk speed is 10000 rpm, 3.5-inch SAS hard disk speed of 15000 rpm, select 2.5-inch hard drive with high IOPS.

As for the server brand, you can choose  IBM, HP or Dell, when the server occupies a high demand for space, you can configure the blade server, such asTecal E6000 Server, 8U space, you can configure the maximum 10 blade server, each server can be equipped with 2 CPUs, 2 SAS hard drives, 12 memory slots, dual port NICs.

Eight, the choice of storage devices

In a virtualized project, it is recommended to use a storage device instead of a server local hard disk, and when you configure a shared storage device, only the virtual machine is saved in storage to quickly implement and use technologies such as Ha, FT, vmotion, and so on. When implementing a virtualization project with VMware vsphere, it is a recommended practice to install VMware ESXi on the local hard disk of the server, This local hard drive can be a solid-state drive (5.2~10GB), or it can be an SD card (configuration 8GB), or even a 1GB USB flash drive, if the server does not configure a local hard disk, you can also partition the server from the storage 8~16GB partitions for booting.

when choosing a storage device, consider the storage capacity, disk performance, number of interfaces, and bandwidth required for the entire virtualized system. for capacity, the capacity of the entire storage design is more than twice times the capacity of actual use, for example, the entire data center has used 1TB of disk space (all the space added together), when the storage is designed, To design at least 2TB of storage space (after configuring raid instead of not configuring raid, adding space to all disks).

For example: If you need 2TB of space, in the use of 600GB hard disk, with RAID10, you need 8 hard disk, the actual capacity is 4 hard disk capacity, 600GBX4≈2.4TB. If you want to use RAID5, you need 5 hard drives.

Another important parameter in the storage design is IOPS (input/output Operations per Second), which is the number of read/write (I/O) operations per second, used in databases and other applications, to measure the performance of random access, IOPS performance on the storage side is different from the host-side Io, which refers to how many hosts per second the storage can receive, and the host's IO requires multiple accesses to the storage to complete. For example, the host writes to a minimum data block, but also through the "Send write request, write data, receive write acknowledgement," such as three steps, that is, 3 storage-side access, each disk system ioPS is capped, if the design of the storage system, the actual IOPS exceeded the upper limit of the disk group, the system response will be slow, Affect the performance of the system. In a nutshell, the ioPS of a 15000-rpm disk is 100 of the ioPS of a 150,10000 disk, and the ioPS of a typical SATA hard disk is about 70~80. In general, when doing desktop virtualization, the IOPS of each virtual machine can be designed to be one of three, the average virtual server IOPS can be planned as 15~30 (see the fact), when designing a system that runs 100 VMs simultaneously, the IOPS should be planned at least 2000, If you use a 10000-RPM SAS disk, you need at least 20 disks, which is a simple measure, and you need to consider a number of factors in real implementation.

When planning storage, also consider the number of interfaces and the speed of the interface, in general, in planning a 4 host, a storage system, with 2 interfaces, 4 SAS interface Storage Server is more appropriate, if there are more hosts, or the host needs redundant interface, You can consider the storage of the FC interface, and use the fiber switch to connect the storage and the server, currently, such as  S2600t, IBM 3524 and other product parameters can meet the requirements of small enterprise virtualization deployment.

Nine, network and switch selection

In a virtualized environment, each physical server generally has a higher network card density, virtualization host has 6, 8 or even more network interface card (NIC) is common, conversely, no virtualized server only 2 or 4 NICs, this becomes a problem in the data center, Because an edge or distribution switch is placed in a rack to simplify network cabling and then transfer up to the network core, a typical 48-port switch can handle only 4~8 virtual hosts, requiring more edge or distribution switches in order to fully fill the rack.

In a virtualized environment, when multiple workloads are consolidated into these hosts, network traffic increases based on the number of workloads running on the host, and network utilization is no longer as low as every physical server in the past.

In order to adjust the increased network traffic from the consolidated workload, you may need to increase the number of upward transfers from the edge or distribution switch to the core of the network, at which point the backplane bandwidth and upstream line of the switch are high.

Another key change comes from the dynamic nature of the latest generation of virtualized products, with such things as hot migration and multi-host dynamic resource management. The dynamic change performance inherent in virtualization means no further assumptions can be made on the flow of traffic between servers.

In the case of a dynamic migration between virtual machines, or when migrating virtual machines from one storage to another, in order to reduce the time of the migration, the impact on critical business is not significant, while the migration will consume a large amount of network resources, in addition, while the migration can reduce the number of concurrent migrations, in some applications, Multiple virtual machines may be migrated at the same time, which requires higher switch backplane bandwidth and the performance of the switch.

In addition, virtualization reduces some of the visibility of the network layer in the data Center, network engineers do not have visibility in the virtual switch, and it is not easy to decide which physical NIC corresponds to which virtual switch, which is the most important information in troubleshooting, in order to reduce the failure rate, Configuring redundant business boards and redundant power supplies for the switch should also be considered, while at the same time, configuring a higher switch with near-possible conditions.

In most cases, the physical host is configured with a 4-port gigabit NIC, and for redundancy, nearly every two NICs are bundled together for load balancing and failover.

For SMB virtualization environments, configuring the S5700 series Gigabit switches for virtualized systems to meet most requirements,  S5700 series 24-Port, 48-Port two, if you need higher network performance, you can choose S9300 Series switch, if in the virtualization planning, Virtual machines in the physical host need only be in the same network segment (or in two finite network segments), and the performance requirements are not high but the price is sensitive, you can choose S1700 series of ordinary switches. Whether VMware ESXi or hyper-V Server, the VLAN is supported in the virtual switch, the host network card is connected to the trunk port of the switch, and then the VLAN is divided on one end of the virtual switch, so that when there are only a two physical network cards, Virtual machines can be partitioned into different VLANs in the owning network.

Figure 6 Model for the switch in the SMB virtualization environment

Enterprise experience sharing with VMware vsphere for server virtualization

Enterprise experience sharing with VMware vsphere for server virtualization