system revision 1.4.2.2
1 1.4.2.2 matt $NetBSD: system,v 1.4.2.2 2008/03/23 00:23:07 matt Exp $ 2 1.4.2.2 matt 3 1.4.2.2 matt NetBSD System Roadmap 4 1.4.2.2 matt ===================== 5 1.4.2.2 matt 6 1.4.2.2 matt This is a small roadmap document, and deals with the main system 7 1.4.2.2 matt aspects of the operating system. 8 1.4.2.2 matt 9 1.4.2.2 matt NetBSD 5.0 will ship with the following main changes to the system: 10 1.4.2.2 matt 11 1.4.2.2 matt 1. Modularized scheduler 12 1.4.2.2 matt 2. Real-time scheduling classes and priorities 13 1.4.2.2 matt 3. Processor sets, processor affinity and processor control 14 1.4.2.2 matt 4. Multiprocessor optimized scheduler 15 1.4.2.2 matt 5. High-performance 1:1 threading implementation 16 1.4.2.2 matt 6. Pushback of the global kernel lock 17 1.4.2.2 matt 7. New kernel concurrency model 18 1.4.2.2 matt 8. Multiprocessor optimized memory allocators 19 1.4.2.2 matt 9. POSIX asynchronous I/O and message queues 20 1.4.2.2 matt 10. In-kernel linker 21 1.4.2.2 matt 11. SysV IPC tuneables 22 1.4.2.2 matt 12. Improved observability: minidumps, lockstat and tprof 23 1.4.2.2 matt 13. Power management framework 24 1.4.2.2 matt 25 1.4.2.2 matt We are working on the following elements, and expect them to be in 5.0: 26 1.4.2.2 matt 27 1.4.2.2 matt 14. Full kernel preemption for real-time threads 28 1.4.2.2 matt 15. 64-bit time values supported 29 1.4.2.2 matt 16. POSIX shared memory 30 1.4.2.2 matt 17. namei() tactical changes 31 1.4.2.2 matt 32 1.4.2.2 matt The following projects are expected to be included in NetBSD 6.0 33 1.4.2.2 matt 34 1.4.2.2 matt 18. Better resource controls 35 1.4.2.2 matt 19. Improved observability: online crashdumps, remote debugging 36 1.4.2.2 matt 20. Processor and cache topology aware scheduler 37 1.4.2.2 matt 38 1.4.2.2 matt We expect to branch for the 5.0 release in April or May 2008, which would 39 1.4.2.2 matt put a NetBSD 5.0 Release date in Q4 2008. An annual release cycle would 40 1.4.2.2 matt thus mean that 6.0 would be branched in April or May 2009. 41 1.4.2.2 matt 42 1.4.2.2 matt Pictorially, this looks like: 43 1.4.2.2 matt 44 1.4.2.2 matt 2008 2009 45 1.4.2.2 matt Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug 46 1.4.2.2 matt |===|===|===|===|===|===|===|===|===|===|===|===|===|===|===|===|===|===|===| 47 1.4.2.2 matt 48 1.4.2.2 matt | ------- 5.0 --------- | | ------- 6.0 49 1.4.2.2 matt branch release branch 50 1.4.2.2 matt 51 1.4.2.2 matt 1-8 11,15-17,18 9,10,12-14 19 20 52 1.4.2.2 matt 53 1.4.2.2 matt We'll continue to update this roadmap as features and dates get firmed up. 54 1.4.2.2 matt 55 1.4.2.2 matt 56 1.4.2.2 matt Some explanations 57 1.4.2.2 matt ================= 58 1.4.2.2 matt 59 1.4.2.2 matt 1. Modularized scheduler 60 1.4.2.2 matt ------------------------ 61 1.4.2.2 matt 62 1.4.2.2 matt Traditionally the only method of control on process scheduling was the 63 1.4.2.2 matt 'nice' value assigned to each process. The scheduler interface has been 64 1.4.2.2 matt redesiged to allow for pluggable schedulers, selected at compile time. 65 1.4.2.2 matt At the current time, there are no plans to switch schedulers at run-time, 66 1.4.2.2 matt since there is little appreciable gain to be had from that, and the extra 67 1.4.2.2 matt performance hit to provide this functionality is thought not to be worth 68 1.4.2.2 matt it. 69 1.4.2.2 matt 70 1.4.2.2 matt The in-kernel scheduler interface has been enhanced to provide a framework 71 1.4.2.2 matt for adding new schedulers, called the common scheduler framework - more 72 1.4.2.2 matt information can be found in the csf(9) manual page. 73 1.4.2.2 matt 74 1.4.2.2 matt Responsible: ad, dsieger, rmind, yamt 75 1.4.2.2 matt 76 1.4.2.2 matt 2. Real-time scheduling classes and priorities 77 1.4.2.2 matt ---------------------------------------------- 78 1.4.2.2 matt 79 1.4.2.2 matt The scheduler has been extended to allow provide multiple new priority 80 1.4.2.2 matt bands, including real-time. POSIX standard interfaces for controlling 81 1.4.2.2 matt thread priority and scheduling class have been implemented, along with 82 1.4.2.2 matt a command line tool to allow control by the system administrator. 83 1.4.2.2 matt 84 1.4.2.2 matt 3. Processor sets, processor affinity and processor control 85 1.4.2.2 matt ----------------------------------------------------------- 86 1.4.2.2 matt 87 1.4.2.2 matt A Solaris and HP-UX compatible interface for defining and controlling 88 1.4.2.2 matt processor sets has been added. Processor sets allow applications and 89 1.4.2.2 matt the administrator complete flexibility in partitioning CPU resources 90 1.4.2.2 matt among applications, down to thread-level granularity. 91 1.4.2.2 matt 92 1.4.2.2 matt Linux compatibile interface controlling processor affinity, similar 93 1.4.2.2 matt in spirit to processor sets, is provided. 94 1.4.2.2 matt 95 1.4.2.2 matt A new utility to control CPU status (cpuctl) is provided. cpuctl 96 1.4.2.2 matt allows the administrator to enable and disable individual CPUs at 97 1.4.2.2 matt the software level, while the system is running. It is expected that 98 1.4.2.2 matt this will in time be extended to support full dynamic reconfiguration, 99 1.4.2.2 matt in concert with a hypervisor such as Xen. 100 1.4.2.2 matt 101 1.4.2.2 matt 4. Multiprocessor optimized scheduler 102 1.4.2.2 matt ------------------------------------- 103 1.4.2.2 matt 104 1.4.2.2 matt An intelligent, pluggable scheduler named M2 that is optimized for 105 1.4.2.2 matt multiprocessor systems, supports POSIX real-time extensions, 106 1.4.2.2 matt time-sharing class, and implements thread affinity. 107 1.4.2.2 matt 108 1.4.2.2 matt 5. High-performance 1:1 threading implementation 109 1.4.2.2 matt ------------------------------------------------ 110 1.4.2.2 matt 111 1.4.2.2 matt A new lightweight 1:1 threading implementation, replacing the M:N based 112 1.4.2.2 matt implementation found in NetBSD 4.0 and earlier. The new implementation is 113 1.4.2.2 matt more correct according to POSIX thread standards, and provides a massive 114 1.4.2.2 matt performance boost to threaded workloads in both uni- and multi-processor 115 1.4.2.2 matt configurations. 116 1.4.2.2 matt 117 1.4.2.2 matt 6. Pushback of the global kernel lock 118 1.4.2.2 matt ------------------------------------- 119 1.4.2.2 matt 120 1.4.2.2 matt Previously, most access to the kernel was single threaded on multiprocessor 121 1.4.2.2 matt systems by the global kernel_lock. The kernel_lock has been pushed back to 122 1.4.2.2 matt to the device driver and wire-protocol layers, providing a significant 123 1.4.2.2 matt performance boost on heavily loaded multiprocessor systems. 124 1.4.2.2 matt 125 1.4.2.2 matt 7. New kernel concurrency model 126 1.4.2.2 matt ------------------------------- 127 1.4.2.2 matt 128 1.4.2.2 matt The non-preemptive spinlock and "interrupt priority level" synchronization 129 1.4.2.2 matt model has been replaced wholesale with a hybrid thread/interrupt model. A 130 1.4.2.2 matt full range of new, lightweight synchronization primitives are available to 131 1.4.2.2 matt the kernel programmer, including: adaptive mutexes, reader/writer locks, 132 1.4.2.2 matt memory barriers, atomic operations, threaded soft interrupts, generic cross 133 1.4.2.2 matt calls, workqueues, priority inheritance, and per-CPU storage. 134 1.4.2.2 matt 135 1.4.2.2 matt 8. Multiprocessor optimized memory allocators 136 1.4.2.2 matt --------------------------------------------- 137 1.4.2.2 matt 138 1.4.2.2 matt The memory allocators in both the kernel and user space are now fully 139 1.4.2.2 matt optimized for multiprocessor systems and eliminate the performance 140 1.4.2.2 matt degradation typically associated with memory allocators in an MP setting. 141 1.4.2.2 matt 142 1.4.2.2 matt 9. POSIX asynchronous I/O and message queues 143 1.4.2.2 matt --------------------------------------------- 144 1.4.2.2 matt 145 1.4.2.2 matt A full implementation of the POSIX asynchronous I/O and message 146 1.4.2.2 matt queue facilities is now available. 147 1.4.2.2 matt 148 1.4.2.2 matt 10. In-kernel linker 149 1.4.2.2 matt -------------------- 150 1.4.2.2 matt 151 1.4.2.2 matt A in-kernel ELF object linker has been added, and a revamped kernel module 152 1.4.2.2 matt infrastructure developed to accompany it. It is expected that the kernel 153 1.4.2.2 matt will become completely modular over time, while continuing to retain the 154 1.4.2.2 matt ability to link to a single binary image for embedded and hobby systems. 155 1.4.2.2 matt 156 1.4.2.2 matt 11. SysV IPC tuneables 157 1.4.2.2 matt ---------------------- 158 1.4.2.2 matt 159 1.4.2.2 matt Parameters for the SVR3-compatible IPC mechanisms can now be tuned 160 1.4.2.2 matt completely at runtime. 161 1.4.2.2 matt 162 1.4.2.2 matt 12. Improved observability: minidumps, lockstat and tprof 163 1.4.2.2 matt --------------------------------------------------------- 164 1.4.2.2 matt 165 1.4.2.2 matt The x86 architecture now supports mini crash-dumps as a support aid for 166 1.4.2.2 matt kernel debugging. Only memory contents actively in use by the kernel at 167 1.4.2.2 matt the time of crash are dumped to and recovered from disk, an improvement 168 1.4.2.2 matt over the traditional scheme where the complete contents of memory is 169 1.4.2.2 matt dumped to disk. 170 1.4.2.2 matt 171 1.4.2.2 matt The lockstat and tprof commands have been addded to the system. lockstat 172 1.4.2.2 matt provides a high-resolution description of lock activity in a running system. 173 1.4.2.2 matt 174 1.4.2.2 matt tprof uses sample based profiling in conjuction with the available 175 1.4.2.2 matt performance counters in order to better profile system activity. 176 1.4.2.2 matt 177 1.4.2.2 matt 13. Power management framework 178 1.4.2.2 matt ------------------------------ 179 1.4.2.2 matt 180 1.4.2.2 matt A new power management framework has been introduced that improves 181 1.4.2.2 matt handling of device power state transitions. As power management support 182 1.4.2.2 matt is now integrated with the auto-configuration subsystem, the kernel can 183 1.4.2.2 matt ensure that a parent device is powered on before attempting to access 184 1.4.2.2 matt the device. 185 1.4.2.2 matt 186 1.4.2.2 matt With these changes comes an updated release of the Intel ACPI 187 1.4.2.2 matt Component Architecture and an x86 emulator which assists in restoring 188 1.4.2.2 matt uninitialized display adapters. 189 1.4.2.2 matt 190 1.4.2.2 matt Leveraging this work, the i386 and amd64 kernels now support suspend 191 1.4.2.2 matt to RAM in uni- and multi-processor configurations on ACPI-capable 192 1.4.2.2 matt machines. This support has been successfully tested on a wide variety of 193 1.4.2.2 matt laptops, including (but not limited to) recent systems from Dell, IBM/Lenovo, 194 1.4.2.2 matt Fujitsu, Toshiba, and Sony. 195 1.4.2.2 matt 196 1.4.2.2 matt Responsible: jmcneill, joerg 197 1.4.2.2 matt 198 1.4.2.2 matt 14. Full kernel preemption for real-time threads 199 1.4.2.2 matt ------------------------------------------------ 200 1.4.2.2 matt 201 1.4.2.2 matt With the revamp of the kernel concurrency model, much of the kernel is fully 202 1.4.2.2 matt multi-threaded and can therefore be preempted at any time. In support of 203 1.4.2.2 matt lower context switch and dispatch times for real-time threads, full kernel 204 1.4.2.2 matt preemption is being implemented. 205 1.4.2.2 matt 206 1.4.2.2 matt 15. 64-bit time_t support 207 1.4.2.2 matt ------------------------- 208 1.4.2.2 matt 209 1.4.2.2 matt The Unix 32-bit time_t value will overflow in 2037 - any mortgage calculations 210 1.4.2.2 matt which use a time_t value are in danger of overflowing at the present time - 211 1.4.2.2 matt and to address this, 64-bit time_t values will be used to contain the number 212 1.4.2.2 matt of seconds since 1970. 213 1.4.2.2 matt 214 1.4.2.2 matt Responsible: christos 215 1.4.2.2 matt 216 1.4.2.2 matt 16. POSIX shared memory 217 1.4.2.2 matt ----------------------- 218 1.4.2.2 matt 219 1.4.2.2 matt Implement POSIX shared memory facilities, which can be used to create the 220 1.4.2.2 matt shared memory objects and add the memory locations to the address space of 221 1.4.2.2 matt a process. 222 1.4.2.2 matt 223 1.4.2.2 matt Responsible: rmind 224 1.4.2.2 matt 225 1.4.2.2 matt 17. Incremental namei improvements, Phase 1 226 1.4.2.2 matt ------------------------------------------- 227 1.4.2.2 matt 228 1.4.2.2 matt Implement the rest of the changes to namei outlined in Message-ID: 229 1.4.2.2 matt <20080319053709.GB3951 (a] netbsd.org>. Simplify the locking and behavior 230 1.4.2.2 matt of namei() calls within the kernel to resolve path names within file 231 1.4.2.2 matt systems. This phase simplifies the majority of calls to namei(). 232 1.4.2.2 matt 233 1.4.2.2 matt Responsible: dholland 234 1.4.2.2 matt 235 1.4.2.2 matt 18. Better resource controls 236 1.4.2.2 matt ---------------------------- 237 1.4.2.2 matt 238 1.4.2.2 matt A resource provisioning and control framework that extends beyond the 239 1.4.2.2 matt traditional Unix process limits. 240 1.4.2.2 matt 241 1.4.2.2 matt 19. Improved observability: online crashdumps, remote debugging 242 1.4.2.2 matt --------------------------------------------------------------- 243 1.4.2.2 matt 244 1.4.2.2 matt XXX crashdumps while the system is running 245 1.4.2.2 matt XXX firewire support in libkvm 246 1.4.2.2 matt 247 1.4.2.2 matt 20. Processor and cache topology aware scheduler 248 1.4.2.2 matt ------------------------------------------------ 249 1.4.2.2 matt 250 1.4.2.2 matt Implement the detection of the topology of the processors and caches. 251 1.4.2.2 matt Improve the scheduler to make decisions about thread migration 252 1.4.2.2 matt according to the topology, to get better thread affinity and less 253 1.4.2.2 matt cache thrashing, and thus improve overall performance in modern SMP 254 1.4.2.2 matt systems. 255 1.4.2.2 matt 256 1.4.2.2 matt Responsible: rmind 257 1.4.2.2 matt 258 1.4.2.2 matt 29. Incremental namei improvements, Phase 2 259 1.4.2.2 matt ------------------------------------------- 260 1.4.2.2 matt 261 1.4.2.2 matt Implement the rest of the changes to namei outlined in Message-ID: 262 1.4.2.2 matt <20080319053709.GB3951 (a] netbsd.org>. Simplify the locking and behavior 263 1.4.2.2 matt of namei() calls within the kernel to resolve path names within file 264 1.4.2.2 matt systems. 265 1.4.2.2 matt 266 1.4.2.2 matt Responsible: dholland 267 1.4.2.2 matt 268 1.4.2.2 matt 269 1.4.2.2 matt 270 1.4.2.2 matt Andrew Doran 271 1.4.2.2 matt Wed Mar 19 18:26:49 GMT 2008 272
Indexes created Sat Nov 01 05:10:12 GMT 2025