Based on Air Max 2012 Black Navy Blue White the virtual 597806-400 Nike LeBron X EXT QS Denim-Pink Outlet machine and use Vagrant (Annotation: Vagrant used to create and deploy automated virtualized development environment) development environment is a major bottleneck in the performance of the file system. CPU difference between the platforms is very small, even negligible, but a lot of RAM and only in the virtual machine (Annotation: refers to virtual machines on the same physical machine) is active only when they become a performance bottleneck. Yesterday I spent most of the time to test and Air Jordan 579756 403 Purple White Nike Black Mamba 24 Kobe Sale Outlet analyze a common file system mechanism, and now I share with you the results. I will begin the analysis of the results, because it is part of the vast majority of people are most interested in. Analysis can be seen behind the other specific testing methods such as use of the software, I get the results of the original 597806-400 Nike LeBron X EXT QS Denim-Pink Outlet data. Each graph presented below, we test different ways to read and write to a file. For each chart, write the total file size is fixed. Y-axis is in KB / s throughput units, X-axis is the 'record size' or one-time read / write large blocks of data size, in kilobytes (KB). Different test environment are as follows: local, VirtualBox local, VMware native, VirtualBox shared folder (vboxsf), VMware shared folder (vmhgfs), NFS. 'Local' refers to a test environment own file system. 'Local' is on the host machine, 'VirtualBox local' is on VirtualBox virtual machine with the device, and so on. NFS only tested on VirtualBox, VirtualBox and VMware because performance characteristics should be very similar. For all icons, throughput (Y-axis) the higher the better. Sequential read small files first, is a serial 64KB file read operation, read various record sizes tested. In a real environment for small files serial read operations are generally loaded application run-time source code, compile, or testing. The first thing you notice, but powerless to change things than, NFS read performance for small files could not believe it. NFS performance is likely to be very tough because due to the large number of read-ahead and caching operations. I for NFS performance than good local virtual file system theory is not well explained. VMware shared folder here just to get rid of VirtualBox shared folder. Read performance VirtualBox shared folder simply terrible. If you've seen the original data, you will never find throughput exceeding 100MB / s, then VMware never less than 500MB / s, even when the peak is 900MB / s. Interestingly, sometimes virtual in the local file system performance even better than the host. This test is original with no user space cache read () system call. Virtual Machine Manager is likely to be carried out ahead of the cache read operation 585388-083 Anti-Nerf Nike KD V Elite Outlet of the virtual machine, so their performance will be the host operating system kernel context switching is better than required. Above theory can also be derived from the raw data call fread () benchmark program. In those tests, the host of the local file system every time to dress up the virtual machine file system (translation: adidas adiPure Crazyquick Collegiate Royal/White Q33301 Outlet one very different fread () and read () is that the former buffer zone, which is without a buffer, so If you have a cushion here, apparently the local file system performance is better). Random read large files where random read throughput test blocks of arbitrary size from 64MB file, and as above is to read a variety of record size. This file is 1000 times before the test file. This type of operation can be seen in the database when processing the read request. Comparative test on a small file for serial reading, the performance difference VMware and VirtualBox Shared Folders shared folder becomes very large. VirtualBox performance was very poor, so that you do not easily see it in the image above. Again, VirtualBox throughput never more than 100MB / s, on the other hand, VMware peak is 7GB / s. Because VirtualBox throughput substantially no difference in various tests, I infer VirtualBox shared folder system there is a code that limits the performance of the thermal path, obviously you do not have a place VirtualBox. NFS does not become prominent until the trial, probably because of its pre-read in the test where the benefit is limited. Nevertheless, compared to other options, NFS still did very well. But, like previous experiments, we still found that the performance is better than the host of the virtual machine, and this is because the Nike Air Max virtual machine manager would be wise to do cache Air Max 2011 Womens Purple Black Grey operation, however, the original host system calls are not It would allow such things to happen. Serial write small files Let's look for a serial writes of small files. This situation is very accurate description of the storage segment status, temporary files, or writing new source code files. The first thing worth noting is that for such a write operation NFS performance becomes very poor. Here we can not do caching for NFS operations, so you have to pay the price because the cost of bringing the network, as well as host there again written to disk, and finally have to wait VM (Virtual Machine) write success signal. Good heavens! Various 'local' file system performance was very good, and again defeated the virtual machine host, once again, this is because the virtual machine manager cache operations. Shared file system performance is similar, but this test is significantly VirtualBox beat VMware. Large file random write us look at the last table is a large random write test file (64MB). Like our large file random read test, this is a good performance for the database how to test. Here is really small files and serial write experiment almost, because there is a large test file, so the difference between different test will be larger, but in addition, the results are basically the same. NFS write performance is still bad time. VirtualBox shared folder on write operations continue beat VMware, and Virtual Machine Manager performance is better than the host. Virtual Machine Manager is a good performance than the host where I am most interested. Test results clearly marked, Virtual Machine Manager must be lied synchronous write (translation: rather than true synchronous write). This also confirms Packer I have seen, if the virtual machine is not properly closed, who writes you submit will be lost. Fsync virtual machine () call is not represented data is written to the hard disk on the host, but the write operation was presented to the virtual Air Max 2011 Womens Grey Green machine manager. Overall analysis placed the shared file system, VMware have what you want Virtual Machine Manager. Loading web pages, run the test case, the Mens Nike Free Run 3 Shoes Black 3 compiler software belong to heavy read. Read performance of VMware shared folder will kill VirtualBox, write performance despite VirtualBox shared folder that is slightly better Nike Lebron 10(X) than VMware. If you choose to use NFS, then use NFS. To emphasize once again, read performance than write performance more valuable. Virtual Machine Manager read / write performance is very good (because it is cheating). Because of these data, I would definitely focus on Vagrant new file system using only local sync folder implementations (such as rsync, or use the local file system as an NFS client Nike Free Run 3 instead of the server). More directly: If you use virtual machines for development, if possible, the database file out of the shared file system. You will see a big performance boost. Finally, I do not think there is very surprising results over the place, since 2010, Vagrant already supports NFS synchronize folders, because we have long realized that the poor performance of the shared folder. However, with some data to indicate different behavior it is worth doing, and this is why each system do provide some interesting insights. 2015 Nike Free 5.0 Test software, configuration, raw data the host is a 2012's Retina MacBook Pro, the hard disk 256G SSD, running Mac OS X10.9.1 operating system, where the cost of land marked. VirtualBox Nike Free Run 3 Women version is 4.3.4, run Ubuntu12.04 virtual machine, which installs and uses the VirtualBox client plug-in. Test equipment is formatted with VirtualBox's city ext3 file system. In with the test device has been marked as VirtualBox locally. However, tests in the shared file system (vboxsf) is marked on a VirtualBox shared folder (vboxsf). VMware Fusion version 6.0.2 running Ubuntu12.04, which install and use the VMware Tools 5.0. VMware tests the device was formatted with ext3 filesystem city. In with the test device has been marked as VMware locally. However, tests in the shared file system (vmhgfs) is marked on a VMware shared folder (vmhgfs). NFS server is shipped with OS X 10.0.1 and constructed. NFS Client from Ubuntu12.04 the nfs-common package, using the NFS protocol over UDP 3, tested on NFS is marked as NFS in VirtualBox. Benchmark software is compiled from Iozone Filesystem Benchmark for 64 from the linux source code. The same binary be used for each test scenario, in addition to a local test is compiled from source from the lozone 32 位 OS X binary. Lozone parameters are -Racb. In addition to the above-mentioned, the remaining default settings and are not mentioned or not used. The original test data above can be obtained in this Excel workbook. There are many other icons in this imgur album. Notes: [1]: This is a standard web development environment, not the typical CPU-intensive or RAM-sensitive. Vagrant wikipedia: http: //en.wikipedia.org/wiki/Vagrant_ (software) Comparative performanceVirtual Machine File System