Debian Codegraph @

UG9/cn/throughput.po

raw
1 # SOME DESCRIPTIVE TITLE
2 # Copyright (C) YEAR Free Software Foundation, Inc.
3 # This file is distributed under the same license as the PACKAGE package.
4 # FIRST AUTHOR <EMAIL@ADDRESS>, YEAR.
5 #
6 #, fuzzy
7 msgid ""
8 msgstr ""
9 "Project-Id-Version: PACKAGE VERSION\n"
10 "POT-Creation-Date: 2020-03-02 15:09+0000\n"
11 "PO-Revision-Date: YEAR-MO-DA HO:MI+ZONE\n"
12 "Last-Translator: FULL NAME <EMAIL@ADDRESS>\n"
13 "Language-Team: LANGUAGE <[email protected]>\n"
14 "Language: \n"
15 "MIME-Version: 1.0\n"
16 "Content-Type: text/plain; charset=CHARSET\n"
17 "Content-Transfer-Encoding: 8bit\n"
18
19 #. type: Title ==
20 #: throughput.adoc:2
21 #, no-wrap
22 msgid "Optimizing DRBD throughput"
23 msgstr "优化DRBD吞吐量"
24
25 #. type: Plain text
26 #: throughput.adoc:7
27 msgid ""
28 "This chapter deals with optimizing DRBD throughput. It examines some "
29 "hardware considerations with regard to throughput optimization, and details "
30 "tuning recommendations for that purpose."
31 msgstr "本章讨论优化DRBD吞吐量。它研究了与吞吐量优化有关的一些硬件考虑因素,并详细介绍了为此目的提出的优化建议。"
32
33 #. type: Title ===
34 #: throughput.adoc:9
35 #, no-wrap
36 msgid "Hardware considerations"
37 msgstr "硬件注意事项"
38
39 #. type: Plain text
40 #: throughput.adoc:14
41 msgid ""
42 "DRBD throughput is affected by both the bandwidth of the underlying I/O "
43 "subsystem (disks, controllers, and corresponding caches), and the bandwidth "
44 "of the replication network."
45 msgstr "DRBD吞吐量受底层I/O子系统(磁盘、控制器和相应缓存)的带宽和复制网络的带宽的影响。"
46
47 #. type: Block title
48 #: throughput.adoc:15
49 #, no-wrap
50 msgid "I/O subsystem throughput"
51 msgstr "I/O子系统吞吐量"
52
53 #. type: Plain text
54 #: throughput.adoc:28
55 msgid ""
56 "indexterm:[throughput]I/O subsystem throughput is determined, largely, by "
57 "the number and type of storage units (disks, SSDs, other Flash storage [like"
58 " FusionIO], ...) that can be written to in parallel. A single, reasonably "
59 "recent, SCSI or SAS disk will typically allow streaming writes of roughly "
60 "40MiB/s to the single disk; an SSD will do 300MiB/s; one of the recent Flash"
61 " storages (NVMe) will be at 1GiB/s. When deployed in a striping "
62 "configuration, the I/O subsystem will parallelize writes across disks, "
63 "effectively multiplying a single disk's throughput by the number of stripes "
64 "in the configuration. Thus the same, 40MB/s disks will allow effective "
65 "throughput of 120MB/s in a RAID-0 or RAID-1+0 configuration with three "
66 "stripes, or 200MB/s with five stripes; with SSDs and/or NVMe you can easily "
67 "get to 1GiB/sec."
68 msgstr ""
69 "indexterm:[throughput]I/O子系统吞吐量在很大程度上取决于可并行写入的存储单元(磁盘、固态硬盘、其他闪存[如FusionIO],...)的数量和类型。一个相当新的SCSI或SAS磁盘通常允许向单个磁盘流式写入大约40MiB/秒;一个固态硬盘将执行300MiB/秒;一个最新的闪存(NVMe)将为1GiB/秒。在分条配置中部署时,I/O子系统将并行化跨磁盘的写入,有效地将单个磁盘的吞吐量乘以配置中的条带数。因此,同样的,40MB/s磁盘将允许有效吞吐量120MB/s的RAID-0或RAID-1+0配置(带三个条带),或200MB/s(带五个条带);使用固态硬盘和/或NVMe,您可以轻松地达到1GiB/s。"
70
71 #. type: Plain text
72 #: throughput.adoc:33
73 msgid ""
74 "A RAID-controller with RAM and a <<s-hardware-bbu,BBU>> can speed up short "
75 "spikes (by buffering them), and so too-short benchmark tests might show "
76 "speeds like 1GiB/s too; for sustained writes its buffers will just run full,"
77 " and then not be of much help, though."
78 msgstr ""
79 "一个带有RAM和<<s-hardware-"
80 "bbu,BBU>>的RAID控制器可以加速短峰值(通过缓冲它们),因此太短的基准测试可能也会显示类似于1GiB/s的速度;对于持续的写操作,它的缓冲区只会满负荷运行,然后就没有多大帮助。"
81
82 #. type: Plain text
83 #: throughput.adoc:39
84 msgid ""
85 "Disk _mirroring_ (RAID-1) in hardware typically has little, if any, effect "
86 "on throughput. Disk _striping with parity_ (RAID-5) does have an effect on "
87 "throughput, usually an adverse one when compared to striping; RAID-5 and "
88 "RAID-6 in software even more so."
89 msgstr ""
90 "硬件中的磁盘镜像(RAID-1)通常对吞吐量的影响很小(如果有的话)。带奇偶校验的磁盘条带(RAID-5)确实对吞吐量有影响,与条带相比通常是不利的;软件中的RAID-5和RAID-6更是如此。"
91
92 #. type: Block title
93 #: throughput.adoc:40
94 #, no-wrap
95 msgid "Network throughput"
96 msgstr "网络吞吐量"
97
98 #. type: Plain text
99 #: throughput.adoc:50
100 msgid ""
101 "indexterm:[throughput]Network throughput is usually determined by the amount"
102 " of traffic present on the network, and on the throughput of any "
103 "routing/switching infrastructure present. These concerns are, however, "
104 "largely irrelevant in DRBD replication links which are normally dedicated, "
105 "back-to-back network connections. Thus, network throughput may be improved "
106 "either by switching to a higher-throughput hardware (such as 10 Gigabit "
107 "Ethernet, or 56GiB Infiniband), or by using link aggregation over several "
108 "network links, as one may do using the Linux indexterm:[bonding "
109 "driver]`bonding` network driver."
110 msgstr ""
111 "indexterm:[throughput]网络吞吐量通常由网络上存在的流量以及存在的任何路由/交换基础设施的吞吐量决定。然而,这些问题在通常是专用的back-to-back网络连接的DRBD复制链路中基本上是不相关的。因此,可以通过切换到更高吞吐量的硬件(例如10 Gigabit以太网或56 GiBInfiniband)或通过使用多个网络链路上的链路聚合来提高网络吞吐量,就像可以使用Linux"
112 " indexterm:[bonding driver]`bonding` 网络驱动程序那样。"
113
114 #. type: Title ===
115 #: throughput.adoc:52
116 #, no-wrap
117 msgid "Throughput overhead expectations"
118 msgstr "吞吐量开销预期"
119
120 #. type: Plain text
121 #: throughput.adoc:56
122 msgid ""
123 "When estimating the throughput overhead associated with DRBD, it is "
124 "important to consider the following natural limitations:"
125 msgstr "在估计与DRBD相关联的吞吐量开销时,必须考虑以下自然限制:"
126
127 #. type: Plain text
128 #: throughput.adoc:58
129 msgid "DRBD throughput is limited by that of the raw I/O subsystem."
130 msgstr "DRBD吞吐量受到原始I/O子系统的限制。"
131
132 #. type: Plain text
133 #: throughput.adoc:59
134 msgid "DRBD throughput is limited by the available network bandwidth."
135 msgstr "DRBD的吞吐量受到可用网络带宽的限制。"
136
137 #. type: Plain text
138 #: throughput.adoc:64
139 msgid ""
140 "The _lower_ of these two establishes the theoretical throughput _maximum_ "
141 "available to DRBD. DRBD then reduces that throughput number by its "
142 "additional overhead, which can be expected to be less than 3 percent."
143 msgstr "这两个值的 _lower_ 建立了DRBD可用的理论吞吐量 _最大值_ 。然后,DRBD通过其额外的开销(可以预期小于3%)来减少吞吐量。"
144
145 #. type: Plain text
146 #: throughput.adoc:71
147 msgid ""
148 "Consider the example of two cluster nodes containing I/O subsystems capable "
149 "of 600 MB/s throughput, with a Gigabit Ethernet link available between them."
150 " Gigabit Ethernet can be expected to produce 110 MB/s throughput for TCP "
151 "connections, thus the network connection would be the bottleneck in this "
152 "configuration and one would expect about 110 MB/s maximum DRBD throughput."
153 msgstr ""
154 "以两个集群节点为例,它们包含能够达到600 "
155 "MB/s吞吐量的I/O子系统,并且在它们之间有一个千兆以太网链路。千兆位以太网可以预期为TCP连接产生110MB/s的吞吐量,因此网络连接将是此配置中的瓶颈,并且可以预期最大DRBD吞吐量约为110 MB/s。"
156
157 #. type: Plain text
158 #: throughput.adoc:75
159 msgid ""
160 "By contrast, if the I/O subsystem is capable of only 80 MB/s for sustained "
161 "writes, then it constitutes the bottleneck, and you should expect only about"
162 " 77 MB/s maximum DRBD throughput."
163 msgstr ""
164 "相比之下,如果I/O子系统仅能够以80 MB/s的速度进行持续的写操作,那么它就构成了瓶颈,您应该只期望大约77 MB/s的最大DRBD吞吐量。"
165
166 #. type: Title ===
167 #: throughput.adoc:78
168 #, no-wrap
169 msgid "Tuning recommendations"
170 msgstr "调整建议"
171
172 #. type: Plain text
173 #: throughput.adoc:88
174 msgid ""
175 "DRBD offers a number of configuration options which may have an effect on "
176 "your system's throughput. This section list some recommendations for tuning "
177 "for throughput. However, since throughput is largely hardware dependent, the"
178 " effects of tweaking the options described here may vary greatly from system"
179 " to system. It is important to understand that these recommendations should "
180 "not be interpreted as \"silver bullets\" which would magically remove any "
181 "and all throughput bottlenecks."
182 msgstr ""
183 "DRBD提供了许多配置选项,这些选项可能会影响系统的吞吐量。本节列出了一些有关调整吞吐量的建议。然而,由于吞吐量在很大程度上依赖于硬件,因此调整此处描述的选项的效果可能因系统而异。重要的是要明白,这些建议不应被解释为能神奇地消除任何和所有吞吐量瓶颈的 `银弹` 。"
184
185 #. type: Title ====
186 #: throughput.adoc:90
187 #, no-wrap
188 msgid "Setting `max-buffers` and `max-epoch-size`"
189 msgstr "设置 `max-buffers` 和 `max-epoch-size` "
190
191 #. type: Plain text
192 #: throughput.adoc:100
193 msgid ""
194 "These options affect write performance on the secondary nodes. `max-buffers`"
195 " is the maximum number of buffers DRBD allocates for writing data to disk "
196 "while `max-epoch-size` is the maximum number of write requests permitted "
197 "between two write barriers. `max-buffers` must be equal or bigger to `max-"
198 "epoch-size` to increase performance. The default for both is 2048; setting "
199 "it to around 8000 should be fine for most reasonably high-performance "
200 "hardware RAID controllers."
201 msgstr ""
202 "这些选项会影响辅助节点上的写入性能。`max buffers` 是DRBD为将数据写入磁盘而分配的最大缓冲区数,而 `max-epoch-"
203 "size` 是两个写入屏障之间允许的最大写入请求数。`max buffers` 必须等于或大于 `max-epoch-"
204 "size` ,才能提高性能。两者的默认值都是2048;对于大多数合理的高性能硬件RAID控制器来说,将其设置为8000左右应该没问题。"
205
206 #. type: delimited block -
207 #: throughput.adoc:111
208 #, no-wrap
209 msgid ""
210 "resource <resource> {\n"
211 " net {\n"
212 " max-buffers 8000;\n"
213 " max-epoch-size 8000;\n"
214 " ...\n"
215 " }\n"
216 " ...\n"
217 "}\n"
218 msgstr ""
219 "resource <resource> {\n"
220 " net {\n"
221 " max-buffers 8000;\n"
222 " max-epoch-size 8000;\n"
223 " ...\n"
224 " }\n"
225 " ...\n"
226 "}\n"
227
228 #. type: Title ====
229 #: throughput.adoc:114
230 #, no-wrap
231 msgid "Tuning the TCP send buffer size"
232 msgstr "调整TCP发送缓冲区大小"
233
234 #. type: Plain text
235 #: throughput.adoc:123
236 msgid ""
237 "The TCP send buffer is a memory buffer for outgoing TCP traffic. By default,"
238 " it is set to a size of 128 KiB. For use in high-throughput networks (such "
239 "as dedicated Gigabit Ethernet or load-balanced bonded connections), it may "
240 "make sense to increase this to a size of 2MiB, or perhaps even more. Send "
241 "buffer sizes of more than 16MiB are generally not recommended (and are also "
242 "unlikely to produce any throughput improvement)."
243 msgstr ""
244 "TCP发送缓冲区是用于传出TCP通信的内存缓冲区。默认情况下,它的大小设置为128 "
245 "KiB。对于在高吞吐量网络(如专用千兆以太网或负载平衡的绑定连接)中使用,将其大小增加到2MiB或更大可能是有意义的。通常不建议发送缓冲区大小超过16MiB(而且也不太可能产生任何吞吐量改进)。"
246
247 #. type: delimited block -
248 #: throughput.adoc:133
249 #, no-wrap
250 msgid ""
251 "resource <resource> {\n"
252 " net {\n"
253 " sndbuf-size 2M;\n"
254 " ...\n"
255 " }\n"
256 " ...\n"
257 "}\n"
258 msgstr ""
259 "resource <resource> {\n"
260 " net {\n"
261 " sndbuf-size 2M;\n"
262 " ...\n"
263 " }\n"
264 " ...\n"
265 "}\n"
266
267 #. type: Plain text
268 #: throughput.adoc:139
269 msgid ""
270 "DRBD also supports TCP send buffer auto-tuning. After enabling this feature,"
271 " DRBD will dynamically select an appropriate TCP send buffer size. TCP send "
272 "buffer auto tuning is enabled by simply setting the buffer size to zero:"
273 msgstr ""
274 "DRBD还支持TCP发送缓冲区自动调整。启用此功能后,DRBD将动态选择适当的TCP发送缓冲区大小。只需将缓冲区大小设置为零即可启用TCP发送缓冲区自动调整:"
275
276 #. type: delimited block -
277 #: throughput.adoc:149
278 #, no-wrap
279 msgid ""
280 "resource <resource> {\n"
281 " net {\n"
282 " sndbuf-size 0;\n"
283 " ...\n"
284 " }\n"
285 " ...\n"
286 "}\n"
287 msgstr ""
288 "resource <resource> {\n"
289 " net {\n"
290 " sndbuf-size 0;\n"
291 " ...\n"
292 " }\n"
293 " ...\n"
294 "}\n"
295
296 #. type: Plain text
297 #: throughput.adoc:155
298 msgid ""
299 "Please note that your ``sysctl``'s settings `net.ipv4.tcp_rmem` and "
300 "`net.ipv4.tcp_wmem` will still influence the behaviour; you should check "
301 "these settings, and perhaps set them similar to `131072 1048576 16777216` "
302 "(minimum 128kiB, default 1MiB, max 16MiB)."
303 msgstr ""
304 "请注意,您的 `sysctl` 设置 `net.ipv4.tcp_rmem` 和 `net.ipv4.tcp_wmem` 仍将影响行为;您应该检查这些设置,并可能将它们设置为类似于 `131072"
305 " 1048576 16777216` (最小128kiB,默认1MiB,最大16MiB)。"
306
307 #. type: Plain text
308 #: throughput.adoc:159
309 msgid ""
310 "`net.ipv4.tcp_mem` is a different beast, with a different unit - do not "
311 "touch, wrong values can easily push your machine into out-of-memory "
312 "situations!"
313 msgstr "`net.ipv4.tcp_mem` 是另一个怪兽,有不同的单位-不要碰,错误的值很容易把你的机器推入内存不足的情况!"
314
315 #. type: Title ====
316 #: throughput.adoc:161
317 #, no-wrap
318 msgid "Tuning the Activity Log size"
319 msgstr "调整活动日志大小"
320
321 #. type: Plain text
322 #: throughput.adoc:167
323 msgid ""
324 "If the application using DRBD is write intensive in the sense that it "
325 "frequently issues small writes scattered across the device, it is usually "
326 "advisable to use a fairly large activity log. Otherwise, frequent metadata "
327 "updates may be detrimental to write performance."
328 msgstr ""
329 "如果使用DRBD的应用程序是写密集型的,因为它经常在设备上分散地发出小的写操作,那么通常建议使用相当大的活动日志。否则,频繁的元数据更新可能会损害写入性能。"
330
331 #. type: delimited block -
332 #: throughput.adoc:177
333 #, no-wrap
334 msgid ""
335 "resource <resource> {\n"
336 " disk {\n"
337 " al-extents 6007;\n"
338 " ...\n"
339 " }\n"
340 " ...\n"
341 "}\n"
342 msgstr ""
343 "resource <resource> {\n"
344 " disk {\n"
345 " al-extents 6007;\n"
346 " ...\n"
347 " }\n"
348 " ...\n"
349 "}\n"
350
351 #. For tuning on striped RAID setups, please see the <<al-stripes,al-stripes>>
352 #. and
353 #. <<al-stripe-size,al-stripe-size>> settings, too.
354 #. type: Title ====
355 #: throughput.adoc:185
356 #, no-wrap
357 msgid "Disabling barriers and disk flushes"
358 msgstr "禁用屏障和磁盘刷新"
359
360 #. type: Plain text
361 #: throughput.adoc:189
362 msgid ""
363 "The recommendations outlined in this section should be applied _only_ to "
364 "systems with non-volatile (battery backed) controller caches."
365 msgstr "本节概述的建议应仅适用于具有非易失性(电池支持)控制器缓存的系统。"
366
367 #. type: Plain text
368 #: throughput.adoc:194
369 msgid ""
370 "Systems equipped with battery backed write cache come with built-in means of"
371 " protecting data in the face of power failure. In that case, it is "
372 "permissible to disable some of DRBD's own safeguards created for the same "
373 "purpose. This may be beneficial in terms of throughput:"
374 msgstr ""
375 "配备电池支持的写缓存的系统配备了内置的方法,可以在断电时保护数据。在这种情况下,允许禁用为相同目的而创建的一些DRBD自己的安全措施。这可能对吞吐量有利:"
376
377 #. type: delimited block -
378 #: throughput.adoc:205
379 #, no-wrap
380 msgid ""
381 "resource <resource> {\n"
382 " disk {\n"
383 " disk-barrier no;\n"
384 " disk-flushes no;\n"
385 " ...\n"
386 " }\n"
387 " ...\n"
388 "}\n"
389 msgstr ""
390 "resource <resource> {\n"
391 " disk {\n"
392 " disk-barrier no;\n"
393 " disk-flushes no;\n"
394 " ...\n"
395 " }\n"
396 " ...\n"
397 "}\n"
398
399 #. type: Title ===
400 #: throughput.adoc:209
401 #, no-wrap
402 msgid "Achieving better Read Performance via increased Redundancy"
403 msgstr "通过增加冗余实现更好的读取性能"
404
405 #. type: Plain text
406 #: throughput.adoc:213
407 msgid ""
408 "As detailed in the man page of `drbd.conf` under `read-balancing`, you can "
409 "increase your read performance by adding more copies of your data."
410 msgstr "如 `read-balancing` 下 `drbd.conf` 的手册页所述,可以通过添加更多数据副本来提高读取性能。"
411
412 #. type: Plain text
413 #: throughput.adoc:217
414 msgid ""
415 "As a ballpark figure: with a single node processing read requests, `fio` on "
416 "a __FusionIO__ card gave us 100k IOPs; after enabling `read-balancing`, the "
417 "performance jumped to 180k IOPs, i.e. +80%!"
418 msgstr ""
419 "一个大概的数字是:单节点处理读请求时,FusionIO卡上的 `fio`给了我们10万IOPs;启用 `读平衡` 后,性能跃升到18万IOPs,即+80%!"
420
421 #. type: Plain text
422 #: throughput.adoc:220
423 msgid ""
424 "So, in case you're running a read-mostly workload (big databases with lots "
425 "of random reads), it might be worth a try to turn `read-balancing` on - and,"
426 " perhaps, add another copy for still more read IO throughput."
427 msgstr ""
428 "因此,如果您运行的是以读为主的工作负载(具有大量随机读操作的大型数据库),那么可能值得尝试启用 `读平衡` ,并且,也许可以添加另一个副本以获得更大的读IO吞吐量。"