Virtual Memory and I/OI/O SystemsTechniques to Improve I/O PerformanceOther Techniques to Improve I/O PerformanceSummary of First PaperSummary of Second PaperFeatures of IO-LiteRelated work before IO-LiteKey Data StructuresDiscussion IImpact of Immutable I/O BuffersDiscussion IIWhat does IO-Lite do?IO-Lite and IPCIo-Lite and FilesystemCopy Semantics Illustration 1Copy Semantics Illustration 2Copy Semantics Illustration 3More on File Cache Management & VM PagingIO-Lite and Network SubsystemA Cross-Subsystem OptimizationPerformance – CompetitorsPerformance – Static Content requestingPerformance – CGI ProgramsPerformance – Real WorkloadPerformance – WAN EffectsPerformance – Other ApplicationsConclusion on I/O-LiteSoftware Prefetching & Caching for TLBsIssues in Virtual MemoryMotivationsApproachDiscussionPrefetching: What entries to prefetch?Prefetching: DetailsPrefetching: PerformanceSlide 37Caching: Software Victim CacheCaching: BenefitsCaching: PerformanceSlide 41Prefetching + Caching: PerformanceSlide 43Virtual Memory and I/OMingsheng HongI/O SystemsMajor I/O HardwareHard disks, network adaptors …Problems related with I/O SystemsVarious types of Hardware – device drivers to provide OS with a unified I/O interfaceTypically much slower than CPU and memory speed – system bottleneckToo much CPU involvement in I/O operationsTechniques to Improve I/O PerformanceBufferinge.g. download a file from networkDMACachingCPU cache, TLB, file cache..Other Techniques to Improve I/O PerformanceVirtual Memory Page Remapping (IO-Lite)Allows (cached) files and memory to be shared by different processes without extra data copyPrefetching Data (Software Pretching and Caching for TLBs)Prefetches and caches page table entriesSummary of First PaperIO-Lite: A Unified I/O Buffering and Caching System (Pai et al. Best Paper of 3rd OSDI, 1999)A unified I/O SystemUses immutable data buffers to store all I/O data (only one physical copy)Uses VM page remappingIPCfile system (disk files, file cache)network subsystemSummary of Second PaperSoftware Prefetching and Caching for Translation Lookaside buffers (Bala et al. 1994)A software approach to help reduce TLB missesWorks well for IPC-intensive systemsBigger performance gain for future systemsFeatures of IO-LiteEliminates redundant data copyingSaves CPU work & avoids cache pollution Eliminates Multiple bufferingSaves main memory => improves hit rate of file cacheEnables cross-subsystem optimizationsCache Internet checksumSupports application-specific cache replacement policiesRelated work before IO-LiteI/O APIs should preserve copy semanticsMemory-mapped filesCopy On WriteFbufsKey Data StructuresImmutable Buffers and Buffer AggregatesDiscussion IWhen we pass a buffer aggregate from process A to process B, how to efficiently do VM page remapping (modify B’s page table entries)?Possible Approach 1: find any empty entry, and modify the VM address contained in buffer aggregateVery inefficientPossible Approach 2: reserve the range of virtual addresses of buffers in the address space of each process Basically limited the total size of buffers – How about dynamically allocated buffers?Impact of Immutable I/O BuffersCopy-On-Write OptimizationModified values are stored in a new buffer, as opposed to “in-place modification”Three situations when the data object is …Completely modifiedAllocates a new bufferPartially modified (modification localized)Chains unmodified and modified portions of dataPartially modified (modification not localized)Compares the cost of writing an entire object with that of chaining; chooses the cheaper methodDiscussion IIHow to measure the two costs?Heuristics neededFragmented data v.s. clustered dataChained data increase reading costSimilar to shadow page technique used in System RShould the cost of retrieving data from buffer also be considered?What does IO-Lite do?Reduces extra data copy in IPCfile system (disk files, file cache)network subsystemMakes possible cross-subsystem optimizationIO-Lite and IPCOperations on Buffers & AggregatesWhen I/O data is transferredPass related aggregates by valueAssociated buffers are passed by referenceWhen buffer is deallocatedBuffer returned to a memory poolBuffer’s VM page mappings persistWhen buffer is reused (by the same process)No further VM map changes required(Temporarily) grant write permission to associated producer processIo-Lite and FilesystemIO-Lite I/O APIs ProvidedIOL_read(int fd, IOL_Agg **aggr, size_t size)IOL_write(int fd, IOL_Agg **aggr)IOL_write operations are atomic – concurrency supportI/O functions in stdio library reimplementedFilesystem cache reorganized Buffer aggregates (pointers to data), instead of file data, are stored in cacheCopy Semantics ensuredSuppose a portion of a cached file is read, and then is overwrittenCopy Semantics Illustration 1Copy Semantics Illustration 2Copy Semantics Illustration 3More on File Cache Management & VM PagingCache replacement policy (can be customized)The eviction order is by current reference status & time of last file accessEvict one entry when the file cache “appears” to be too largeAdded one entry on every file cache missWhen a buffer page is paged out, data will be written back to swap space, and possibly to several other disk locations (for different files)IO-Lite and Network SubsystemAccess control and protection for processesACL related with buffer poolsMust determine the ACL of a data object prior to allocating memory for itEarly demultiplexing technique to determine ACL for each incoming packetA Cross-Subsystem OptimizationInternet checksum cachingCache the computed checksum for each slice of a buffer aggregateIncrement the version number when buffer is reallocated – can be used to check whether data changed Works well for static files. Also has a big benefit on the CGI programs that chain dynamic data with static dataPerformance – CompetitorsFlash Web server – a high performance HTTP serverFlash-Lite – A modified version of Flash using IO-Lite APIApache 1.3.1 – representing the widely used Web server todayPerformance – Static Content requestingPerformance – CGI ProgramsPerformance – Real