4 This document describes RCU-related publications, and is followed by
5 the corresponding bibtex entries.
7 The first thing resembling RCU was published in 1980, when Kung and Lehman
8 [Kung80] recommended use of a garbage collector to defer destruction
9 of nodes in a parallel binary search tree in order to simplify its
10 implementation. This works well in environments that have garbage
11 collectors, but current production garbage collectors incur significant
14 In 1982, Manber and Ladner [Manber82,Manber84] recommended deferring
15 destruction until all threads running at that time have terminated, again
16 for a parallel binary search tree. This approach works well in systems
17 with short-lived threads, such as the K42 research operating system.
18 However, Linux has long-lived tasks, so more is needed.
20 In 1986, Hennessy, Osisek, and Seigh [Hennessy89] introduced passive
21 serialization, which is an RCU-like mechanism that relies on the presence
22 of "quiescent states" in the VM/XA hypervisor that are guaranteed not
23 to be referencing the data structure. However, this mechanism was not
24 optimized for modern computer systems, which is not surprising given
25 that these overheads were not so expensive in the mid-80s. Nonetheless,
26 passive serialization appears to be the first deferred-destruction
27 mechanism to be used in production. Furthermore, the relevant patent has
28 lapsed, so this approach may be used in non-GPL software, if desired.
29 (In contrast, use of RCU is permitted only in software licensed under
32 In 1990, Pugh [Pugh90] noted that explicitly tracking which threads
33 were reading a given data structure permitted deferred free to operate
34 in the presence of non-terminating threads. However, this explicit
35 tracking imposes significant read-side overhead, which is undesirable
36 in read-mostly situations. This algorithm does take pains to avoid
37 write-side contention and parallelize the other write-side overheads by
38 providing a fine-grained locking design, however, it would be interesting
39 to see how much of the performance advantage reported in 1990 remains
42 At about this same time, Adams [Adams91] described ``chaotic relaxation'',
43 where the normal barriers between successive iterations of convergent
44 numerical algorithms are relaxed, so that iteration $n$ might use
45 data from iteration $n-1$ or even $n-2$. This introduces error,
46 which typically slows convergence and thus increases the number of
47 iterations required. However, this increase is sometimes more than made
48 up for by a reduction in the number of expensive barrier operations,
49 which are otherwise required to synchronize the threads at the end
50 of each iteration. Unfortunately, chaotic relaxation requires highly
51 structured data, such as the matrices used in scientific programs, and
52 is thus inapplicable to most data structures in operating-system kernels.
54 In 1993, Jacobson [Jacobson93] verbally described what is perhaps the
55 simplest deferred-free technique: simply waiting a fixed amount of time
56 before freeing blocks awaiting deferred free. Jacobson did not describe
57 any write-side changes he might have made in this work using SGI's Irix
58 kernel. Aju John published a similar technique in 1995 [AjuJohn95].
59 This works well if there is a well-defined upper bound on the length of
60 time that reading threads can hold references, as there might well be in
61 hard real-time systems. However, if this time is exceeded, perhaps due
62 to preemption, excessive interrupts, or larger-than-anticipated load,
63 memory corruption can ensue, with no reasonable means of diagnosis.
64 Jacobson's technique is therefore inappropriate for use in production
65 operating-system kernels, except when such kernels can provide hard
66 real-time response guarantees for all operations.
68 Also in 1995, Pu et al. [Pu95a] applied a technique similar to that of Pugh's
69 read-side-tracking to permit replugging of algorithms within a commercial
70 Unix operating system. However, this replugging permitted only a single
71 reader at a time. The following year, this same group of researchers
72 extended their technique to allow for multiple readers [Cowan96a].
73 Their approach requires memory barriers (and thus pipeline stalls),
74 but reduces memory latency, contention, and locking overheads.
76 1995 also saw the first publication of DYNIX/ptx's RCU mechanism
77 [Slingwine95], which was optimized for modern CPU architectures,
78 and was successfully applied to a number of situations within the
79 DYNIX/ptx kernel. The corresponding conference paper appeared in 1998
82 In 1999, the Tornado and K42 groups described their "generations"
83 mechanism, which quite similar to RCU [Gamsa99]. These operating systems
84 made pervasive use of RCU in place of "existence locks", which greatly
85 simplifies locking hierarchies.
87 2001 saw the first RCU presentation involving Linux [McKenney01a]
88 at OLS. The resulting abundance of RCU patches was presented the
89 following year [McKenney02a], and use of RCU in dcache was first
90 described that same year [Linder02a].
92 Also in 2002, Michael [Michael02b,Michael02a] presented techniques
93 that defer the destruction of data structures to simplify non-blocking
94 synchronization (wait-free synchronization, lock-free synchronization,
95 and obstruction-free synchronization are all examples of non-blocking
96 synchronization). In particular, this technique eliminates locking,
97 reduces contention, reduces memory latency for readers, and parallelizes
98 pipeline stalls and memory latency for writers. However, these
99 techniques still impose significant read-side overhead in the form of
100 memory barriers. Researchers at Sun worked along similar lines in the
101 same timeframe [HerlihyLM02,HerlihyLMS03].
103 In 2003, the K42 group described how RCU could be used to create
104 hot-pluggable implementations of operating-system functions. Later that
105 year saw a paper describing an RCU implementation of System V IPC
106 [Arcangeli03], and an introduction to RCU in Linux Journal [McKenney03a].
108 2004 has seen a Linux-Journal article on use of RCU in dcache
109 [McKenney04a], a performance comparison of locking to RCU on several
110 different CPUs [McKenney04b], a dissertation describing use of RCU in a
111 number of operating-system kernels [PaulEdwardMcKenneyPhD], a paper
112 describing how to make RCU safe for soft-realtime applications [Sarma04c],
113 and a paper describing SELinux performance with RCU [JamesMorris04b].
119 ,author="H. T. Kung and Q. Lehman"
120 ,title="Concurrent Maintenance of Binary Search Trees"
123 ,journal="ACM Transactions on Database Systems"
130 ,author="Udi Manber and Richard E. Ladner"
131 ,title="Concurrency Control in a Dynamic Search Structure"
132 ,institution="Department of Computer Science, University of Washington"
133 ,address="Seattle, Washington"
141 ,author="Udi Manber and Richard E. Ladner"
142 ,title="Concurrency Control in a Dynamic Search Structure"
145 ,journal="ACM Transactions on Database Systems"
151 @techreport{Hennessy89
152 ,author="James P. Hennessy and Damian L. Osisek and Joseph W. {Seigh II}"
153 ,title="Passive Serialization in a Multitasking Environment"
154 ,institution="US Patent and Trademark Office"
155 ,address="Washington, DC"
157 ,number="US Patent 4,809,168 (lapsed)"
163 ,author="William Pugh"
164 ,title="Concurrent Maintenance of Skip Lists"
165 ,institution="Institute of Advanced Computer Science Studies, Department of Computer Science, University of Maryland"
166 ,address="College Park, Maryland"
168 ,number="CS-TR-2222.1"
173 ,Author="Gregory R. Adams"
174 ,title="Concurrent Programming, Principles, and Practices"
175 ,Publisher="Benjamin Cummins"
179 @unpublished{Jacobson93
180 ,author="Van Jacobson"
181 ,title="Avoid Read-Side Locking Via Delayed Free"
184 ,note="Verbal discussion"
187 @Conference{AjuJohn95
189 ,Title="Dynamic vnodes -- Design and Implementation"
190 ,Booktitle="{USENIX Winter 1995}"
191 ,Publisher="USENIX Association"
195 ,Address="New Orleans, LA"
198 @techreport{Slingwine95
199 ,author="John D. Slingwine and Paul E. McKenney"
200 ,title="Apparatus and Method for Achieving Reduced Overhead Mutual
201 Exclusion and Maintaining Coherency in a Multiprocessor System
202 Utilizing Execution History and Thread Monitoring"
203 ,institution="US Patent and Trademark Office"
204 ,address="Washington, DC"
206 ,number="US Patent 5,442,758 (contributed under GPL)"
210 @techreport{Slingwine97
211 ,author="John D. Slingwine and Paul E. McKenney"
212 ,title="Method for maintaining data coherency using thread
213 activity summaries in a multicomputer system"
214 ,institution="US Patent and Trademark Office"
215 ,address="Washington, DC"
217 ,number="US Patent 5,608,893 (contributed under GPL)"
221 @techreport{Slingwine98
222 ,author="John D. Slingwine and Paul E. McKenney"
223 ,title="Apparatus and method for achieving reduced overhead
224 mutual exclusion and maintaining coherency in a multiprocessor
225 system utilizing execution history and thread monitoring"
226 ,institution="US Patent and Trademark Office"
227 ,address="Washington, DC"
229 ,number="US Patent 5,727,209 (contributed under GPL)"
233 @Conference{McKenney98
234 ,Author="Paul E. McKenney and John D. Slingwine"
235 ,Title="Read-Copy Update: Using Execution History to Solve Concurrency
237 ,Booktitle="{Parallel and Distributed Computing and Systems}"
241 ,Address="Las Vegas, NV"
245 ,Author="Ben Gamsa and Orran Krieger and Jonathan Appavoo and Michael Stumm"
246 ,Title="Tornado: Maximizing Locality and Concurrency in a Shared Memory
247 Multiprocessor Operating System"
248 ,Booktitle="{Proceedings of the 3\textsuperscript{rd} Symposium on
249 Operating System Design and Implementation}"
253 ,Address="New Orleans, LA"
256 @techreport{Slingwine01
257 ,author="John D. Slingwine and Paul E. McKenney"
258 ,title="Apparatus and method for achieving reduced overhead
259 mutual exclusion and maintaining coherency in a multiprocessor
260 system utilizing execution history and thread monitoring"
261 ,institution="US Patent and Trademark Office"
262 ,address="Washington, DC"
264 ,number="US Patent 5,219,690 (contributed under GPL)"
268 @Conference{McKenney01a
269 ,Author="Paul E. McKenney and Jonathan Appavoo and Andi Kleen and
270 Orran Krieger and Rusty Russell and Dipankar Sarma and Maneesh Soni"
271 ,Title="Read-Copy Update"
272 ,Booktitle="{Ottawa Linux Symposium}"
276 \url{http://www.linuxsymposium.org/2001/abstracts/readcopy.php}
277 \url{http://www.rdrop.com/users/paulmck/rclock/rclock_OLS.2001.05.01c.pdf}
278 [Viewed June 23, 2004]"
280 Described RCU, and presented some patches implementing and using it in
285 @Conference{Linder02a
286 ,Author="Hanna Linder and Dipankar Sarma and Maneesh Soni"
287 ,Title="Scalability of the Directory Entry Cache"
288 ,Booktitle="{Ottawa Linux Symposium}"
294 @Conference{McKenney02a
295 ,Author="Paul E. McKenney and Dipankar Sarma and
296 Andrea Arcangeli and Andi Kleen and Orran Krieger and Rusty Russell"
297 ,Title="Read-Copy Update"
298 ,Booktitle="{Ottawa Linux Symposium}"
303 \url{http://www.linux.org.uk/~ajh/ols2002_proceedings.pdf.gz}
304 [Viewed June 23, 2004]"
308 ,author="J. Appavoo and K. Hui and C. A. N. Soules and R. W. Wisniewski and
309 D. M. {Da Silva} and O. Krieger and M. A. Auslander and D. J. Edelsohn and
310 B. Gamsa and G. R. Ganger and P. McKenney and M. Ostrowski and
311 B. Rosenburg and M. Stumm and J. Xenidis"
312 ,title="Enabling Autonomic Behavior in Systems Software With Hot Swapping"
315 ,journal="IBM Systems Journal"
321 @Conference{Arcangeli03
322 ,Author="Andrea Arcangeli and Mingming Cao and Paul E. McKenney and
324 ,Title="Using Read-Copy Update Techniques for {System V IPC} in the
326 ,Booktitle="Proceedings of the 2003 USENIX Annual Technical Conference
328 ,Publisher="USENIX Association"
335 ,author="Paul E. McKenney"
336 ,title="Using {RCU} in the {Linux} 2.5 Kernel"
339 ,journal="Linux Journal"
345 @techreport{Friedberg03a
346 ,author="Stuart A. Friedberg"
347 ,title="Lock-Free Wild Card Search Data Structure and Method"
348 ,institution="US Patent and Trademark Office"
349 ,address="Washington, DC"
351 ,number="US Patent 6,662,184 (contributed under GPL)"
357 ,author="Paul E. McKenney and Dipankar Sarma and Maneesh Soni"
358 ,title="Scaling dcache with {RCU}"
361 ,journal="Linux Journal"
367 @Conference{McKenney04b
368 ,Author="Paul E. McKenney"
369 ,Title="{RCU} vs. Locking Performance on Different {CPUs}"
370 ,Booktitle="{linux.conf.au}"
373 ,Address="Adelaide, Australia"
375 \url{http://www.linux.org.au/conf/2004/abstracts.html#90}
376 \url{http://www.rdrop.com/users/paulmck/rclock/lockperf.2004.01.17a.pdf}
377 [Viewed June 23, 2004]"
380 @phdthesis{PaulEdwardMcKenneyPhD
381 ,author="Paul E. McKenney"
382 ,title="Exploiting Deferred Destruction:
383 An Analysis of Read-Copy-Update Techniques
384 in Operating System Kernels"
385 ,school="OGI School of Science and Engineering at
386 Oregon Health and Sciences University"
389 \url{http://www.rdrop.com/users/paulmck/RCU/RCUdissertation.2004.07.14e1.pdf}
390 [Viewed October 15, 2004]"
394 ,Author="Dipankar Sarma and Paul E. McKenney"
395 ,Title="Making RCU Safe for Deep Sub-Millisecond Response Realtime Applications"
396 ,Booktitle="Proceedings of the 2004 USENIX Annual Technical Conference
398 ,Publisher="USENIX Association"
404 @unpublished{JamesMorris04b
405 ,Author="James Morris"
406 ,Title="Recent Developments in {SELinux} Kernel Performance"
410 \url{http://www.livejournal.com/users/james_morris/2153.html}
411 [Viewed December 10, 2004]"