1 .\" Copyright (C) 2001 Matthew Dillon. All rights reserved.
3 .\" Redistribution and use in source and binary forms, with or without
4 .\" modification, are permitted provided that the following conditions
6 .\" 1. Redistributions of source code must retain the above copyright
7 .\" notice, this list of conditions and the following disclaimer.
8 .\" 2. Redistributions in binary form must reproduce the above copyright
9 .\" notice, this list of conditions and the following disclaimer in the
10 .\" documentation and/or other materials provided with the distribution.
12 .\" THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
13 .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
14 .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
15 .\" ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
16 .\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
17 .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
18 .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
19 .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
20 .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
21 .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 .Nd performance tuning under FreeBSD
32 .Sh SYSTEM SETUP - DISKLABEL, NEWFS, TUNEFS, SWAP
37 to lay out your file systems on a hard disk it is important to remember
38 that hard drives can transfer data much more quickly from outer tracks
39 than they can from inner tracks.
40 To take advantage of this you should
41 try to pack your smaller file systems and swap closer to the outer tracks,
42 follow with the larger file systems, and end with the largest file systems.
43 It is also important to size system standard file systems such that you
44 will not be forced to resize them later as you scale the machine up.
45 I usually create, in order, a 128M root, 1G swap, 128M
51 and use any remaining space for
54 You should typically size your swap space to approximately 2x main memory
55 for systems with less than 2GB of RAM, or approximately 1x main memory
57 If you do not have a lot of RAM, though, you will generally want a lot
59 It is not recommended that you configure any less than
60 256M of swap on a system and you should keep in mind future memory
61 expansion when sizing the swap partition.
62 The kernel's VM paging algorithms are tuned to perform best when there is
63 at least 2x swap versus main memory.
64 Configuring too little swap can lead
65 to inefficiencies in the VM page scanning code as well as create issues
66 later on if you add more memory to your machine.
67 Finally, on larger systems
68 with multiple SCSI disks (or multiple IDE disks operating on different
69 controllers), we strongly recommend that you configure swap on each drive.
70 The swap partitions on the drives should be approximately the same size.
71 The kernel can handle arbitrary sizes but
72 internal data structures scale to 4 times the largest swap partition.
74 the swap partitions near the same size will allow the kernel to optimally
75 stripe swap space across the N disks.
76 Do not worry about overdoing it a
77 little, swap space is the saving grace of
79 and even if you do not normally use much swap, it can give you more time to
80 recover from a runaway program before being forced to reboot.
84 partition depends heavily on what you intend to use the machine for.
86 partition is primarily used to hold mailboxes, the print spool, and log
90 its own partition (but except for extreme cases it is not worth the waste
92 If your machine is intended to act as a mail
94 or you are running a heavily visited web server, you should consider
95 creating a much larger partition \(en perhaps a gig or more.
97 to underestimate log file storage requirements.
101 depends on the kind of temporary file usage you think you will need.
103 the minimum we recommend.
104 Also note that sysinstall will create a
107 Dedicating a partition for temporary file storage is important for
108 two reasons: first, it reduces the possibility of file system corruption
109 in a crash, and second it reduces the chance of a runaway process that
111 .Oo Pa /var Oc Ns Pa /tmp
112 from blowing up more critical subsystems (mail,
115 .Oo Pa /var Oc Ns Pa /tmp
116 is a very common problem to have.
118 In the old days there were differences between
122 but the introduction of
126 led to massive confusion
127 by program writers so today programs haphazardly use one or the
128 other and thus no real distinction can be made between the two.
129 So it makes sense to have just one temporary directory and
130 softlink to it from the other
135 the one thing you do not want to do is leave it sitting
136 on the root partition where it might cause root to fill up or possibly
137 corrupt root in a crash/reboot situation.
141 partition holds the bulk of the files required to support the system and
142 a subdirectory within it called
144 holds the bulk of the files installed from the
147 If you do not use ports all that much and do not intend to keep
150 on the machine, you can get away with
154 However, if you install a lot of ports
155 (especially window managers and Linux-emulated binaries), we recommend
156 at least a 2 gigabyte
158 and if you also intend to keep system source
159 on the machine, we recommend a 3 gigabyte
161 Do not underestimate the
162 amount of space you will need in this partition, it can creep up and
167 partition is typically used to hold user-specific data.
168 I usually size it to the remainder of the disk.
170 Why partition at all?
171 Why not create one big
173 partition and be done with it?
174 Then I do not have to worry about undersizing things!
175 Well, there are several reasons this is not a good idea.
177 each partition has different operational characteristics and separating them
178 allows the file system to tune itself to those characteristics.
182 partitions are read-mostly, with very little writing, while
183 a lot of reading and writing could occur in
188 partitioning your system fragmentation introduced in the smaller more
189 heavily write-loaded partitions will not bleed over into the mostly-read
191 Additionally, keeping the write-loaded partitions closer to
192 the edge of the disk (i.e., before the really big partitions instead of after
193 in the partition table) will increase I/O performance in the partitions
194 where you need it the most.
195 Now it is true that you might also need I/O
196 performance in the larger partitions, but they are so large that shifting
197 them more towards the edge of the disk will not lead to a significant
198 performance improvement whereas moving
200 to the edge can have a huge impact.
201 Finally, there are safety concerns.
202 Having a small neat root partition that
203 is essentially read-only gives it a greater chance of surviving a bad crash
206 Properly partitioning your system also allows you to tune
213 requires more experience but can lead to significant improvements in
215 There are three parameters that are relatively safe to tune:
216 .Em blocksize , bytes/i-node ,
218 .Em cylinders/group .
221 performs best when using 8K or 16K file system block sizes.
222 The default file system block size is 16K,
223 which provides best performance for most applications,
224 with the exception of those that perform random access on large files
225 (such as database server software).
226 Such applications tend to perform better with a smaller block size,
227 although modern disk characteristics are such that the performance
228 gain from using a smaller block size may not be worth consideration.
229 Using a block size larger than 16K
230 can cause fragmentation of the buffer cache and
231 lead to lower performance.
233 The defaults may be unsuitable
234 for a file system that requires a very large number of i-nodes
235 or is intended to hold a large number of very small files.
236 Such a file system should be created with an 8K or 4K block size.
237 This also requires you to specify a smaller
239 We recommend always using a fragment size that is 1/8
240 the block size (less testing has been done on other fragment size factors).
243 options for this would be
244 .Dq Li "newfs -f 1024 -b 8192 ..." .
246 If a large partition is intended to be used to hold fewer, larger files, such
247 as database files, you can increase the
249 ratio which reduces the number of i-nodes (maximum number of files and
250 directories that can be created) for that partition.
251 Decreasing the number
252 of i-nodes in a file system can greatly reduce
254 recovery times after a crash.
255 Do not use this option
256 unless you are actually storing large files on the partition, because if you
257 overcompensate you can wind up with a file system that has lots of free
258 space remaining but cannot accommodate any more files.
259 Using 32768, 65536, or 262144 bytes/i-node is recommended.
260 You can go higher but
261 it will have only incremental effects on
265 .Dq Li "newfs -i 32768 ..." .
268 may be used to further tune a file system.
269 This command can be run in
270 single-user mode without having to reformat the file system.
271 However, this is possibly the most abused program in the system.
272 Many people attempt to
273 increase available file system space by setting the min-free percentage to 0.
274 This can lead to severe file system fragmentation and we do not recommend
278 option worthwhile here is turning on
281 .Dq Li "tunefs -n enable /filesystem" .
284 and later, softupdates can be turned on using the
290 will typically enable softupdates automatically for non-root file systems).
291 Softupdates drastically improves meta-data performance, mainly file
292 creation and deletion.
293 We recommend enabling softupdates on most file systems; however, there
294 are two limitations to softupdates that you should be aware of when
295 determining whether to use it on a file system.
296 First, softupdates guarantees file system consistency in the
297 case of a crash but could very easily be several seconds (even a minute!\&)
298 behind on pending write to the physical disk.
299 If you crash you may lose more work
301 Secondly, softupdates delays the freeing of file system
303 If you have a file system (such as the root file system) which is
304 close to full, doing a major update of it, e.g.\&
305 .Dq Li "make installworld" ,
306 can run it out of space and cause the update to fail.
307 For this reason, softupdates will not be enabled on the root file system
308 during a typical install.
309 There is no loss of performance since the root
310 file system is rarely written to.
314 options exist that can help you tune the system.
315 The most obvious and most dangerous one is
317 Only use this option in conjunction with
319 as it is far too dangerous on a normal file system.
320 A less dangerous and more
326 file systems normally update the last-accessed time of a file or
327 directory whenever it is accessed.
328 This operation is handled in
330 with a delayed write and normally does not create a burden on the system.
331 However, if your system is accessing a huge number of files on a continuing
332 basis the buffer cache can wind up getting polluted with atime updates,
333 creating a burden on the system.
334 For example, if you are running a heavily
335 loaded web site, or a news server with lots of readers, you might want to
336 consider turning off atime updates on your larger partitions with this
339 However, you should not gratuitously turn off atime
343 file system customarily
344 holds mailboxes, and atime (in combination with mtime) is used to
345 determine whether a mailbox has new mail.
346 You might as well leave
347 atime turned on for mostly read-only partitions such as
352 This is especially useful for
354 since some system utilities
355 use the atime field for reporting.
357 In larger systems you can stripe partitions from several drives together
358 to create a much larger overall partition.
359 Striping can also improve
360 the performance of a file system by splitting I/O operations across two
367 utilities may be used to create simple striped file systems.
369 speaking, striping smaller partitions such as the root and
371 or essentially read-only partitions such as
373 is a complete waste of time.
374 You should only stripe partitions that require serious I/O performance,
377 or custom partitions used to hold databases and web pages.
378 Choosing the proper stripe size is also
380 File systems tend to store meta-data on power-of-2 boundaries
381 and you usually want to reduce seeking rather than increase seeking.
383 means you want to use a large off-center stripe size such as 1152 sectors
384 so sequential I/O does not seek both disks and so meta-data is distributed
385 across both disks rather than concentrated on a single disk.
387 you really need to get sophisticated, we recommend using a real hardware
388 RAID controller from the list of
390 supported controllers.
393 variables permit system behavior to be monitored and controlled at
395 Some sysctls simply report on the behavior of the system; others allow
396 the system behavior to be modified;
397 some may be set at boot time using
399 but most will be set via
401 There are several hundred sysctls in the system, including many that appear
402 to be candidates for tuning but actually are not.
403 In this document we will only cover the ones that have the greatest effect
408 sysctl defines the overcommit behaviour of the vm subsystem.
409 The virtual memory system always does accounting of the swap space
410 reservation, both total for system and per-user.
412 are available through sysctl
414 that gives the total bytes available for swapping, and
415 .Va vm.swap_reserved ,
416 that gives number of bytes that may be needed to back all currently
417 allocated anonymous memory.
421 sysctl causes the virtual memory system to return failure
422 to the process when allocation of memory causes
426 Bit 1 of the sysctl enforces
431 Root is exempt from this limit.
432 Bit 2 allows to count most of the physical
433 memory as allocatable, except wired and free reserved pages
435 .Va vm.stats.vm.v_free_target
437 .Va vm.stats.vm.v_wire_count
438 sysctls, respectively).
441 .Va kern.ipc.maxpipekva
442 loader tunable is used to set a hard limit on the
443 amount of kernel address space allocated to mapping of pipe buffers.
444 Use of the mapping allows the kernel to eliminate a copy of the
445 data from writer address space into the kernel, directly copying
446 the content of mapped buffer to the reader.
447 Increasing this value to a higher setting, such as `25165824' might
448 improve performance on systems where space for mapping pipe buffers
449 is quickly exhausted.
450 This exhaustion is not fatal; however, and it will only cause pipes to
451 to fall back to using double-copy.
454 .Va kern.ipc.shm_use_phys
455 sysctl defaults to 0 (off) and may be set to 0 (off) or 1 (on).
457 this parameter to 1 will cause all System V shared memory segments to be
458 mapped to unpageable physical RAM.
459 This feature only has an effect if you
460 are either (A) mapping small amounts of shared memory across many (hundreds)
461 of processes, or (B) mapping large amounts of shared memory across any
463 This feature allows the kernel to remove a great deal
464 of internal memory management page-tracking overhead at the cost of wiring
465 the shared memory into core, making it unswappable.
468 .Va vfs.vmiodirenable
469 sysctl defaults to 1 (on).
470 This parameter controls how directories are cached
472 Most directories are small and use but a single fragment
473 (typically 2K) in the file system and even less (typically 512 bytes) in
475 However, when operating in the default mode the buffer
476 cache will only cache a fixed number of directories even if you have a huge
478 Turning on this sysctl allows the buffer cache to use
479 the VM Page Cache to cache the directories.
480 The advantage is that all of
481 memory is now available for caching directories.
482 The disadvantage is that
483 the minimum in-core memory used to cache a directory is the physical page
484 size (typically 4K) rather than 512 bytes.
485 We recommend turning this option off in memory-constrained environments;
486 however, when on, it will substantially improve the performance of services
487 that manipulate a large number of files.
488 Such services can include web caches, large mail systems, and news systems.
489 Turning on this option will generally not reduce performance even with the
490 wasted memory but you should experiment to find out.
494 sysctl defaults to 1 (on).
495 This tells the file system to issue media
496 writes as full clusters are collected, which typically occurs when writing
497 large sequential files.
498 The idea is to avoid saturating the buffer
499 cache with dirty buffers when it would not benefit I/O performance.
501 this may stall processes and under certain circumstances you may wish to turn
505 .Va vfs.hirunningspace
506 sysctl determines how much outstanding write I/O may be queued to
507 disk controllers system-wide at any given time.
508 It is used by the UFS file system.
509 The default is self-tuned and
510 usually sufficient but on machines with advanced controllers and lots
511 of disks this may be tuned up to match what the controllers buffer.
512 Configuring this setting to match tagged queuing capabilities of
513 controllers or drives with average IO size used in production works
514 best (for example: 16 MiB will use 128 tags with IO requests of 128 KiB).
515 Note that setting too high a value
516 (exceeding the buffer cache's write threshold) can lead to extremely
517 bad clustering performance.
518 Do not set this value arbitrarily high!
519 Higher write queueing values may also add latency to reads occurring at
524 sysctl governs VFS read-ahead and is expressed as the number of blocks
525 to pre-read if the heuristics algorithm decides that the reads are
527 It is used by the UFS, ext2fs and msdosfs file systems.
528 With the default UFS block size of 16 KiB, a setting of 32 will allow
529 speculatively reading up to 512 KiB.
530 This setting may be increased to get around disk I/O latencies, especially
531 where these latencies are large such as in virtual machine emulated
533 It may be tuned down in specific cases where the I/O load is such that
534 read-ahead adversely affects performance or where system memory is really
539 sysctl defines how large VFS namecache may grow.
540 The number of currently allocated entries in namecache is provided by
542 sysctl and the condition
543 debug.numcache < kern.maxvnodes * vfs.ncsizefactor
548 sysctl defines how many negative entries VFS namecache is allowed to create.
549 The number of currently allocated negative entries is provided by
551 sysctl and the condition
552 vfs.ncnegfactor * debug.numneg < debug.numcache
555 There are various other buffer-cache and VM page cache related sysctls.
556 We do not recommend modifying these values.
559 the VM system does an extremely good job tuning itself.
562 .Va net.inet.tcp.sendspace
564 .Va net.inet.tcp.recvspace
565 sysctls are of particular interest if you are running network intensive
567 They control the amount of send and receive buffer space
568 allowed for any given TCP connection.
569 The default sending buffer is 32K; the default receiving buffer
572 improve bandwidth utilization by increasing the default at the cost of
573 eating up more kernel memory for each connection.
575 increasing the defaults if you are serving hundreds or thousands of
576 simultaneous connections because it is possible to quickly run the system
577 out of memory due to stalled connections building up.
579 high bandwidth over a fewer number of connections, especially if you have
580 gigabit Ethernet, increasing these defaults can make a huge difference.
581 You can adjust the buffer size for incoming and outgoing data separately.
582 For example, if your machine is primarily doing web serving you may want
583 to decrease the recvspace in order to be able to increase the
584 sendspace without eating too much kernel memory.
585 Note that the routing table (see
587 can be used to introduce route-specific send and receive buffer size
590 As an additional management tool you can use pipes in your
593 to limit the bandwidth going to or from particular IP blocks or ports.
594 For example, if you have a T1 you might want to limit your web traffic
595 to 70% of the T1's bandwidth in order to leave the remainder available
596 for mail and interactive use.
597 Normally a heavily loaded web server
598 will not introduce significant latencies into other services even if
599 the network link is maxed out, but enforcing a limit can smooth things
600 out and lead to longer term stability.
601 Many people also enforce artificial
602 bandwidth limitations in order to ensure that they are not charged for
603 using too much bandwidth.
605 Setting the send or receive TCP buffer to values larger than 65535 will result
606 in a marginal performance improvement unless both hosts support the window
607 scaling extension of the TCP protocol, which is controlled by the
608 .Va net.inet.tcp.rfc1323
610 These extensions should be enabled and the TCP buffer size should be set
611 to a value larger than 65536 in order to obtain good performance from
612 certain types of network links; specifically, gigabit WAN links and
613 high-latency satellite links.
614 RFC1323 support is enabled by default.
617 .Va net.inet.tcp.always_keepalive
618 sysctl determines whether or not the TCP implementation should attempt
619 to detect dead TCP connections by intermittently delivering
622 By default, this is enabled for all applications; by setting this
623 sysctl to 0, only applications that specifically request keepalives
625 In most environments, TCP keepalives will improve the management of
626 system state by expiring dead TCP connections, particularly for
627 systems serving dialup users who may not always terminate individual
628 TCP connections before disconnecting from the network.
629 However, in some environments, temporary network outages may be
630 incorrectly identified as dead sessions, resulting in unexpectedly
631 terminated TCP connections.
632 In such environments, setting the sysctl to 0 may reduce the occurrence of
633 TCP session disconnections.
636 .Va net.inet.tcp.delayed_ack
637 TCP feature is largely misunderstood.
638 Historically speaking, this feature
639 was designed to allow the acknowledgement to transmitted data to be returned
640 along with the response.
641 For example, when you type over a remote shell,
642 the acknowledgement to the character you send can be returned along with the
643 data representing the echo of the character.
644 With delayed acks turned off,
645 the acknowledgement may be sent in its own packet, before the remote service
646 has a chance to echo the data it just received.
647 This same concept also
648 applies to any interactive protocol (e.g.\& SMTP, WWW, POP3), and can cut the
649 number of tiny packets flowing across the network in half.
652 delayed ACK implementation also follows the TCP protocol rule that
653 at least every other packet be acknowledged even if the standard 100ms
654 timeout has not yet passed.
655 Normally the worst a delayed ACK can do is
656 slightly delay the teardown of a connection, or slightly delay the ramp-up
657 of a slow-start TCP connection.
658 While we are not sure we believe that
659 the several FAQs related to packages such as SAMBA and SQUID which advise
660 turning off delayed acks may be referring to the slow-start issue.
663 it would be more beneficial to increase the slow-start flightsize via
665 .Va net.inet.tcp.slowstart_flightsize
666 sysctl rather than disable delayed acks.
669 .Va net.inet.tcp.inflight.enable
670 sysctl turns on bandwidth delay product limiting for all TCP connections.
671 The system will attempt to calculate the bandwidth delay product for each
672 connection and limit the amount of data queued to the network to just the
673 amount required to maintain optimum throughput.
674 This feature is useful
675 if you are serving data over modems, GigE, or high speed WAN links (or
676 any other link with a high bandwidth*delay product), especially if you are
677 also using window scaling or have configured a large send window.
678 If you enable this option, you should also be sure to set
679 .Va net.inet.tcp.inflight.debug
680 to 0 (disable debugging), and for production use setting
681 .Va net.inet.tcp.inflight.min
682 to at least 6144 may be beneficial.
683 Note however, that setting high
684 minimums may effectively disable bandwidth limiting depending on the link.
685 The limiting feature reduces the amount of data built up in intermediate
686 router and switch packet queues as well as reduces the amount of data built
687 up in the local host's interface queue.
688 With fewer packets queued up,
689 interactive connections, especially over slow modems, will also be able
690 to operate with lower round trip times.
691 However, note that this feature
692 only affects data transmission (uploading / server-side).
694 affect data reception (downloading).
697 .Va net.inet.tcp.inflight.stab
699 This parameter defaults to 20, representing 2 maximal packets added
700 to the bandwidth delay product window calculation.
702 window is required to stabilize the algorithm and improve responsiveness
703 to changing conditions, but it can also result in higher ping times
704 over slow links (though still much lower than you would get without
705 the inflight algorithm).
706 In such cases you may
707 wish to try reducing this parameter to 15, 10, or 5, and you may also
709 .Va net.inet.tcp.inflight.min
710 (for example, to 3500) to get the desired effect.
711 Reducing these parameters
712 should be done as a last resort only.
715 .Va net.inet.ip.portrange.*
716 sysctls control the port number ranges automatically bound to TCP and UDP
718 There are three ranges: a low range, a default range, and a
719 high range, selectable via the
724 network programs use the default range which is controlled by
725 .Va net.inet.ip.portrange.first
727 .Va net.inet.ip.portrange.last ,
728 which default to 49152 and 65535, respectively.
729 Bound port ranges are
730 used for outgoing connections, and it is possible to run the system out
731 of ports under certain circumstances.
732 This most commonly occurs when you are
733 running a heavily loaded web proxy.
734 The port range is not an issue
735 when running a server which handles mainly incoming connections, such as a
736 normal web server, or has a limited number of outgoing connections, such
738 For situations where you may run out of ports,
739 we recommend decreasing
740 .Va net.inet.ip.portrange.first
742 A range of 10000 to 30000 ports may be reasonable.
743 You should also consider firewall effects when changing the port range.
745 may block large ranges of ports (usually low-numbered ports) and expect systems
746 to use higher ranges of ports for outgoing connections.
748 .Va net.inet.ip.portrange.last
749 is set at the maximum allowable port number.
752 .Va kern.ipc.somaxconn
753 sysctl limits the size of the listen queue for accepting new TCP connections.
754 The default value of 128 is typically too low for robust handling of new
755 connections in a heavily loaded web server environment.
756 For such environments,
757 we recommend increasing this value to 1024 or higher.
759 may itself limit the listen queue size (e.g.\&
762 often have a directive in its configuration file to adjust the queue size up.
763 Larger listen queues also do a better job of fending off denial of service
768 sysctl determines how many open files the system supports.
770 typically a few thousand but you may need to bump this up to ten or twenty
771 thousand if you are running databases or large descriptor-heavy daemons.
774 sysctl may be interrogated to determine the current number of open files
778 .Va vm.swap_idle_enabled
779 sysctl is useful in large multi-user systems where you have lots of users
780 entering and leaving the system and lots of idle processes.
782 tend to generate a great deal of continuous pressure on free memory reserves.
783 Turning this feature on and adjusting the swapout hysteresis (in idle
785 .Va vm.swap_idle_threshold1
787 .Va vm.swap_idle_threshold2
788 allows you to depress the priority of pages associated with idle processes
789 more quickly then the normal pageout algorithm.
790 This gives a helping hand
791 to the pageout daemon.
792 Do not turn this option on unless you need it,
793 because the tradeoff you are making is to essentially pre-page memory sooner
794 rather than later, eating more swap and disk bandwidth.
796 this option will have a detrimental effect but in a large system that is
797 already doing moderate paging this option allows the VM system to stage
798 whole processes into and out of memory more easily.
800 Some aspects of the system behavior may not be tunable at runtime because
801 memory allocations they perform must occur early in the boot process.
802 To change loader tunables, you must set their values in
804 and reboot the system.
807 controls the scaling of a number of static system tables, including defaults
808 for the maximum number of open files, sizing of network memory resources, etc.
812 is automatically sized at boot based on the amount of memory available in
813 the system, and may be determined at run-time by inspecting the value of the
817 Some sites will require larger or smaller values of
819 and may set it as a loader tunable; values of 64, 128, and 256 are not
821 We do not recommend going above 256 unless you need a huge number
822 of file descriptors; many of the tunable values set to their defaults by
824 may be individually overridden at boot-time or run-time as described
825 elsewhere in this document.
828 must set this value via the kernel
838 tunables set the default soft limits for process data and stack size
840 Processes may increase these up to the hard limits by calling
847 tunables set the hard limits for process data, stack, and text size
848 respectively; processes may not exceed these limits.
851 tunable controls how much the stack segment will grow when a process
852 needs to allocate more stack.
854 .Va kern.ipc.nmbclusters
855 may be adjusted to increase the number of network mbufs the system is
857 Each cluster represents approximately 2K of memory,
858 so a value of 1024 represents 2M of kernel memory reserved for network
860 You can do a simple calculation to figure out how many you need.
861 If you have a web server which maxes out at 1000 simultaneous connections,
862 and each connection eats a 16K receive and 16K send buffer, you need
863 approximately 32MB worth of network buffers to deal with it.
865 thumb is to multiply by 2, so 32MBx2 = 64MB/2K = 32768.
867 you would want to set
868 .Va kern.ipc.nmbclusters
870 We recommend values between
871 1024 and 4096 for machines with moderates amount of memory, and between 4096
872 and 32768 for machines with greater amounts of memory.
873 Under no circumstances
874 should you specify an arbitrarily high value for this parameter, it could
875 lead to a boot-time crash.
880 may be used to observe network cluster use.
883 do not have this tunable and require that the
890 More and more programs are using the
892 system call to transmit files over the network.
895 sysctl controls the number of file system buffers
897 is allowed to use to perform its work.
898 This parameter nominally scales
901 so you should not need to modify this parameter except under extreme
907 manual page for details.
908 .Sh KERNEL CONFIG TUNING
909 There are a number of kernel options that you may have to fiddle with in
910 a large-scale system.
911 In order to change these options you need to be
912 able to compile a new kernel from source.
915 manual page and the handbook are good starting points for learning how to
917 Generally the first thing you do when creating your own custom
918 kernel is to strip out all the drivers and services you do not use.
921 and drivers you do not have will reduce the size of your kernel, sometimes
922 by a megabyte or more, leaving more memory available for applications.
925 may be used to reduce system boot times.
926 The defaults are fairly high and
927 can be responsible for 5+ seconds of delay in the boot process.
930 to something below 5 seconds could work (especially with modern drives).
932 There are a number of
934 options that can be commented out.
935 If you only want the kernel to run
936 on a Pentium class CPU, you can easily remove
940 if you are sure your CPU is being recognized as a Pentium II or better.
941 Some clones may be recognized as a Pentium or even a 486 and not be able
942 to boot without those options.
945 will be able to better use higher-end CPU features for MMU, task switching,
946 timebase, and even device operations.
947 Additionally, higher-end CPUs support
948 4MB MMU pages, which the kernel uses to map the kernel itself into memory,
949 increasing its efficiency under heavy syscall loads.
950 .Sh IDE WRITE CACHING
952 flirted with turning off IDE write caching.
953 This reduced write bandwidth
954 to IDE disks but was considered necessary due to serious data consistency
955 issues introduced by hard drive vendors.
956 Basically the problem is that
957 IDE drives lie about when a write completes.
958 With IDE write caching turned
959 on, IDE hard drives will not only write data to disk out of order, they
960 will sometimes delay some of the blocks indefinitely under heavy disk
962 A crash or power failure can result in serious file system
964 So our default was changed to be safe.
966 result was such a huge loss in performance that we caved in and changed the
967 default back to on after the release.
968 You should check the default on
969 your system by observing the
972 If IDE write caching is turned off, you can turn it back
976 More information on tuning the ATA driver system may be found in the
979 If you need performance, go with SCSI.
980 .Sh CPU, MEMORY, DISK, NETWORK
981 The type of tuning you do depends heavily on where your system begins to
982 bottleneck as load increases.
983 If your system runs out of CPU (idle times
984 are perpetually 0%) then you need to consider upgrading the CPU or moving to
985 an SMP motherboard (multiple CPU's), or perhaps you need to revisit the
986 programs that are causing the load and try to optimize them.
988 is paging to swap a lot you need to consider adding more memory.
990 system is saturating the disk you typically see high CPU idle times and
991 total disk saturation.
993 can be used to monitor this.
994 There are many solutions to saturated disks:
995 increasing memory for caching, mirroring disks, distributing operations across
996 several machines, and so forth.
997 If disk performance is an issue and you
998 are using IDE drives, switching to SCSI can help a great deal.
1000 IDE drives compare with SCSI in raw sequential bandwidth, the moment you
1001 start seeking around the disk SCSI drives usually win.
1003 Finally, you might run out of network suds.
1004 The first line of defense for
1005 improving network performance is to make sure you are using switches instead
1006 of hubs, especially these days where switches are almost as cheap.
1008 have severe problems under heavy loads due to collision back-off and one bad
1009 host can severely degrade the entire LAN.
1010 Second, optimize the network path
1011 as much as possible.
1014 we describe a firewall protecting internal hosts with a topology where
1015 the externally visible hosts are not routed through it.
1017 than 10BaseT, or use 1000BaseT rather than 100BaseT, depending on your needs.
1018 Most bottlenecks occur at the WAN link (e.g.\&
1019 modem, T1, DSL, whatever).
1020 If expanding the link is not an option it may be possible to use the
1022 feature to implement peak shaving or other forms of traffic shaping to
1023 prevent the overloaded service (such as web services) from affecting other
1024 services (such as email), or vice versa.
1025 In home installations this could
1026 be used to give interactive traffic (your browser,
1029 over services you export from your box (web services, email).
1063 manual page was originally written by
projects / FreeBSD / releng / 9.0.git / blob