| blob | 76b0f79ca87789099e840cb12162c6d8a20c85a0 |
1 Frontswap provides a "transcendent memory" interface for swap pages.
2 In some environments, dramatic performance savings may be obtained because
3 swapped pages are saved in RAM (or a RAM-like device) instead of a swap disk.
5 Frontswap is so named because it can be thought of as the opposite of
6 a "backing" store for a swap device. The storage is assumed to be
7 a synchronous concurrency-safe page-oriented "pseudo-RAM device" conforming
8 to the requirements of transcendent memory (such as Xen's "tmem", or
9 in-kernel compressed memory, aka "zcache", or future RAM-like devices);
10 this pseudo-RAM device is not directly accessible or addressable by the
11 kernel and is of unknown and possibly time-varying size. The "device"
12 links itself to frontswap by calling frontswap_register_ops to set the
13 frontswap_ops funcs appropriately and the functions it provides must
14 conform to certain policies as follows:
16 An "init" prepares the device to receive frontswap pages associated
17 with the specified swap device number (aka "type"). A "put_page" will
18 copy the page to transcendent memory and associate it with the type and
19 offset associated with the page. A "get_page" will copy the page, if found,
20 from transcendent memory into kernel memory, but will NOT remove the page
21 from from transcendent memory. A "flush_page" will remove the page from
22 transcendent memory and a "flush_area" will remove ALL pages associated
23 with the swap type (e.g., like swapoff) and notify the "device" to refuse
24 further puts with that swap type.
26 Once a page is successfully put, a matching get on the page will always
27 succeed. So when the kernel finds itself in a situation where it needs
28 to swap out a page, it first attempts to use frontswap. If the put returns
29 non-zero, the data has been successfully saved to transcendent memory and
30 a disk write and, if the data is later read back, a disk read are avoided.
31 If a put returns zero, transcendent memory has rejected the data, and the
32 page can be written to swap as usual.
34 Note that if a page is put and the page already exists in transcendent memory
35 (a "duplicate" put), either the put succeeds and the data is overwritten,
36 or the put fails AND the page is flushed. This ensures stale data may
37 never be obtained from psuedo-RAM.
39 Monitoring and control of frontswap is done by sysfs files in the
40 /sys/kernel/mm/frontswap directory. The effectiveness of frontswap can
41 be measured (across all swap devices) with:
43 curr_pages - number of pages currently contained in frontswap
44 failed_puts - how many put attempts have failed
45 gets - how many gets were attempted (all should succeed)
46 succ_puts - how many put attempts have succeeded
47 flushes - how many flushes were attempted
49 The number can be reduced by root by writing an integer target to curr_pages,
50 which results in a "partial swapoff", thus reducing the number of frontswap
51 pages to that target if memory constraints permit.
53 FAQ
55 1) Where's the value?
57 When a workload starts swapping, performance falls through the floor.
58 Frontswap significantly increases performance in many such workloads by
59 providing a clean, dynamic interface to read and write swap pages to
60 "transcendent" memory that is otherwise not directly addressable to the kernel.
61 This interface is ideal when data is transformed to a different form
62 and size (such as with compression) or secretly moved (as might be
63 useful for write-balancing for some RAM-like devices). Swap pages (and
64 evicted page-cache pages) are a great use for this kind of slower-than-RAM-
65 but-much-faster-than-disk "pseudo-RAM device" and the frontswap (and
66 cleancache) interface to transcendent memory provides a nice way to read
67 and write -- and indirectly "name" -- the pages.
69 In the virtual case, the whole point of virtualization is to statistically
70 multiplex physical resources acrosst the varying demands of multiple
71 virtual machines. This is really hard to do with RAM and efforts to do
72 it well with no kernel changes have essentially failed (except in some
73 well-publicized special-case workloads). Frontswap -- and cleancache --
74 with a fairly small impact on the kernel, provides a huge amount
75 of flexibility for more dynamic, flexible RAM multiplexing.
76 Specifically, the Xen Transcendent Memory backend allows otherwise
77 "fallow" hypervisor-owned RAM to not only be "time-shared" between multiple
78 virtual machines, but the pages can be compressed and deduplicated to
79 optimize RAM utilization. And when guest OS's are induced to surrender
80 underutilized RAM (e.g. with "self-ballooning"), sudden unexpected
81 memory pressure may result in swapping; frontswap allows those pages
82 to be swapped to and from hypervisor RAM if overall host system memory
83 conditions allow.
85 2) Sure there may be performance advantages in some situations, but
86 what's the space/time overhead of frontswap?
88 If CONFIG_FRONTSWAP is disabled, every frontswap hook compiles into
89 nothingness and the only overhead is a few extra bytes per swapon'ed
90 swap device. If CONFIG_FRONTSWAP is enabled but no frontswap "backend"
91 registers, there is one extra global variable compared to zero for
92 every swap page read or written. If CONFIG_FRONTSWAP is enabled
93 AND a frontswap backend registers AND the backend fails every "put"
94 request (i.e. provides no memory despite claiming it might),
95 CPU overhead is still negligible -- and since every frontswap fail
96 precedes a swap page write-to-disk, the system is highly likely
97 to be I/O bound and using a small fraction of a percent of a CPU
98 will be irrelevant anyway.
100 As for space, if CONFIG_FRONTSWAP is enabled AND a frontswap backend
101 registers, one bit is allocated for every swap page for every swap
102 device that is swapon'd. This is added to the EIGHT bits (which
103 was sixteen until about 2.6.34) that the kernel already allocates
104 for every swap page for every swap device that is swapon'd. (Hugh
105 Dickins has observed that frontswap could probably steal one of
106 the existing eight bits, but let's worry about that minor optimization
107 later.) For very large swap disks (which are rare) on a standard
108 4K pagesize, this is 1MB per 32GB swap.
110 3) OK, how about a quick overview of what this frontswap patch does
111 in terms that a kernel hacker can grok?
113 Let's assume that a frontswap "backend" has registered during
114 kernel initialization; this registration indicates that this
115 frontswap backend has access to some "memory" that is not directly
116 accessible by the kernel. Exactly how much memory it provides is
117 entirely dynamic and random.
119 Whenever a swap-device is swapon'd frontswap_init() is called,
120 passing the swap device number (aka "type") as a parameter.
121 This notifies frontswap to expect attempts to "put" swap pages
122 associated with that number.
124 Whenever the swap subsystem is readying a page to write to a swap
125 device (c.f swap_writepage()), frontswap_put_page is called. Frontswap
126 consults with the frontswap backend and if the backend says
127 it does NOT have room, frontswap_put_page returns 0 and the page is
128 swapped as normal. Note that the response from the frontswap
129 backend is essentially random; it may choose to never accept a
130 page, it could accept every ninth page, or it might accept every
131 page. But if the backend does accept a page, the data from the page
132 has already been copied and associated with the type and offset,
133 and the backend guarantees the persistence of the data. In this case,
134 frontswap sets a bit in the "frontswap_map" for the swap device
135 corresponding to the page offset on the swap device to which it would
136 otherwise have written the data.
138 When the swap subsystem needs to swap-in a page (swap_readpage()),
139 it first calls frontswap_get_page() which checks the frontswap_map to
140 see if the page was earlier accepted by the frontswap backend. If
141 it was, the page of data is filled from the frontswap backend and
142 the swap-in is complete. If not, the normal swap-in code is
143 executed to obtain the page of data from the real swap device.
145 So every time the frontswap backend accepts a page, a swap device read
146 and (potentially) a swap device write are replaced by a "frontswap backend
147 put" and (possibly) a "frontswap backend get", which are presumably much
148 faster.
150 4) Can't frontswap be configured as a "special" swap device that is
151 just higher priority than any real swap device (e.g. like zswap)?
153 No. Recall that acceptance of any swap page by the frontswap
154 backend is entirely unpredictable. This is critical to the definition
155 of frontswap because it grants completely dynamic discretion to the
156 backend. But since any "put" might fail, there must always be a real
157 slot on a real swap device to swap the page. Thus frontswap must be
158 implemented as a "shadow" to every swapon'd device with the potential
159 capability of holding every page that the swap device might have held
160 and the possibility that it might hold no pages at all.
161 On the downside, this also means that frontswap cannot contain more
162 pages than the total of swapon'd swap devices. For example, if NO
163 swap device is configured on some installation, frontswap is useless.
165 Further, frontswap is entirely synchronous whereas a real swap
166 device is, by definition, asynchronous and uses block I/O. The
167 block I/O layer is not only unnecessary, but may perform "optimizations"
168 that are inappropriate for a RAM-oriented device including delaying
169 the write of some pages for a significant amount of time.
170 Synchrony is required to ensure the dynamicity of the backend.
172 In a virtualized environment, the dynamicity allows the hypervisor
173 (or host OS) to do "intelligent overcommit". For example, it can
174 choose to accept pages only until host-swapping might be imminent,
175 then force guests to do their own swapping.
177 5) Why this weird definition about "duplicate puts"? If a page
178 has been previously successfully put, can't it always be
179 successfully overwritten?
181 Nearly always it can, but no, sometimes it cannot. Consider an example
182 where data is compressed and the original 4K page has been compressed
183 to 1K. Now an attempt is made to overwrite the page with data that
184 is non-compressible and so would take the entire 4K. But the backend
185 has no more space. In this case, the put must be rejected. Whenever
186 frontswap rejects a put that would overwrite, it also must flush
187 the old data and ensure that it is no longer accessible. Since the
188 swap subsystem then writes the new data to the read swap device,
189 this is the correct course of action to ensure coherency.
191 6) What is frontswap_shrink for?
193 When the (non-frontswap) swap subsystem swaps out a page to a real
194 swap device, that page is only taking up low-value pre-allocated disk
195 space. But if frontswap has placed a page in transcendent memory, that
196 page may be taking up valuable real estate. The frontswap_shrink
197 routine allows a process outside of the swap subsystem (such as
198 a userland service via the sysfs interface, or a kernel thread)
199 to force pages out of the memory managed by frontswap and back into
200 kernel-addressable memory.
202 7) Why does the frontswap patch create the new include file swapfile.h?
204 The frontswap code depends on some swap-subsystem-internal data
205 structures that have, over the years, moved back and forth between
206 static and global. This seemed a reasonable compromise: Define
207 them as global but declare them in a new include file that isn't
208 included by the large number of source files that include swap.h.
210 Dan Magenheimer, last updated May 27, 2011