CSR-PDOS: Reduce, Reuse, Recycle: Operating System Support for Server Appliance Consolidation

各种规模的组织一直在购买越来越多的单功能计算设备，如防火墙、邮件过滤器、Web缓存等。随着商用处理器和硬件的能力不断增强，这些设备通常是服务器级PC，其操作系统和软件已预先配置为与硬件匹配。虽然这种方法最初提供了一些简单性，但它的长期问题可能包括需要单独更新操作系统、更难升级以及无法从服务器整合中受益的计算机的激增。此研究调查了在没有完全硬件隔离的情况下操作系统对设备式服务器开发的支持。研究的一些机制涉及硬件辅助的虚拟机、使用OS上下文支持的“软”虚拟机、细粒度的资源隔离和记账，以及更明确的资源管理接口。这样的操作系统支持减少了每台服务器的资源消耗，并提供了在以前隔离的设备之间共享公共基础设施的机制。因此，开发人员仍然可以保留预先调整的操作系统和服务器配置的好处，而不会产生设备共享部分的维护费用。这项研究将对操作系统进行修改，以帮助整合服务器设备。我们将修改服务器以使用这些功能，以证明它们可以减少资源消耗，变得更可预测，并且在此过程中变得更容易编程。我们将发布修改后的操作系统，并使用以前在不同硬件平台上运行的合适服务器进行演示。随着处理器变得更快，核心数量增加，这项研究将简化计算基础设施。

Organizations of all sizes have been purchasing an increasing numberof single-function computing appliances, such as firewalls, mailfilters, Web caches, etc. With the increasing power of commodityprocessors and hardware, these appliances are often server-class PCswith the operating system and software pre-configured to match thehardware. While this approach provides some initial simplicity, itslong-term problems can include a proliferation of machines thatrequire individual OS updates, are harder to upgrade, and that cannotbenefit from server consolidation.This research investigates operating system support forappliance-style server development without complete hardwareisolation. Some of the mechanisms investigated involvehardware-assisted virtual machines, "soft" virtual machines using OScontext support, fine-grained resource isolation and accounting, andmore explicit resource management interfaces. Such OS support reducesthe resource consumption footprints of each server, and providesmechanisms to share common infrastructure across previously isolatedappliances. As a result, developers still retain the benefit ofpre-tuned OS and server configurations, without incurring themaintenance overheads for the shared portions of the appliance.This research will produce operating systems modifications to aidconsolidating server appliances. We will modify servers to use thesefeatures, to demonstrate that they can reduce their resourceconsumption, become more predictable, and also become easier toprogram in the process. We will release the modified operating systemand demonstrate it with suitable servers that previously would havebeen run on separate hardware platforms. As processors become faster,and the number of cores increases, this research will simplifycomputing infrastructure.