NetBSD下日志文件系统和soft deps性能比较

cnhnln

WAPBL vs. Soft Dependences


A few weeks ago I posted about NetBSD’s new journaling technology WAPBL. I received a lot of questions concerning the benchmark results that I published. People were especially interested in a comparison with FFS using soft dependences. I hadn’t used this mode in my original benchmark because my personal experiences with soft dependences weren’t very positive.
I decided nevertheless to repeat the benchmark using the default mode (no mount options), asynchronous mode (-o async), soft dependences (-o softdep) and WAPBL (-o log). I also changed the benchmark itself:

• Before each benchmark run the file system is initialized using a block size of 16KB and a fragment size of 2KB. As no space is reserved at the end of partition WAPBL has to use an in file system journal.
• During the tar phase the benchmark extracts the NetBSD 4.0 source tar archives ten times into seperate directories. The file set is now considerably larger than the main memory of the test machine. The source tar archives were fetched from a memory file system before this phase to make sure that the buffer cache is pre-populated.
• During the rm phase the benchmark removes the 10 directories it created previously.
• Both phases are finished by running the sync command and unmounting the file system to make sure that all buffers have been written to disk.
• The times quoted in the table below include the time it took to unmount the file system. The values were calculated by taking the average of four runs of the benchmark. All values are denoted in minutes:seconds.

Benchmark	synchronous	soft deps	WAPBL	asynchronous
tar	155:56	39:06	35:52	34:07
rm	59:55	9:11	4:36	4:18

Let’s have a look at the results:

• The synchronous mode (synchronous metadata writes) provides the worst performance, the asynchronous mode the best. Anything else would have been a big suprise.
• Soft dependences provide much better performance than the synchronous mode. But they can be considerably slower than the asynchronous mode depending on the work load.
• WAPBL is slower than the asynchronous mode but not by much. It is always faster than soft dependences.

Considering these benchmark results and the fact that only WAPBL can recover from a system crash without an expensive file system check it is the clear winner, at least in my humble opinion.

[ 本帖最后由 cnhnln 于 2008-8-24 19:47 编辑 ]

cnhnln

Journaling performance


Mathias Scheler has an interesting blog posting about the difference of using journaling on file system performance. The test he did was extracting NetBSD 4.0 sources.

• With plain FFS, the extract took 15:19 minutes
• With journaling, it took 3:24 minutes.

A clear winner. (No numbers with soft dependencies, though, but they can be expected to be comparable to the journaling number).

[ 本帖最后由 cnhnln 于 2008-8-24 19:47 编辑 ]

cnhnln

Journaling at last


A wish came true: NetBSD finally has a journaling file system. Simon Burge added support for WAPBL, Write Ahead Physical Block Logging file system journaling, to NetBSD-current’s main branch today.
The main purpose of a journaling file system is to avoid a time consuming file system checks after a power failure, system crash or similar problem. But it can also help to improve performance considerably depending on the design. As I was curious about WAPBL’s performance impact I ran my usual benchmark (extracting the NetBSD 4.0 source tar archives). And the result is very encouraging: FFS required 15:19 minutes to finish the benchmark without logging and only 3:24 minutes to finish it with logging. It seems that WAPBL provides similar speed improvements as soft dependencies and keeps your file system safe at the same time.
A big thank you to Wasabi Systems for donating the code and to Simon Burge and all the other developers for integrating WAPBL into NetBSD!

[ 本帖最后由 cnhnln 于 2008-8-24 19:48 编辑 ]

cnhnln

Trying out journaling  
After NetBSD got journaling integrated into FFS recently, I've built and installed -current, and had a look. In short: it works just as expected. In other works: Yai! :-) :-) :-)
The wapbl(4) manpage gives more details: To enable, a kernel with "options WAPBL" needs to run, which is available in NetBSD-current since end of July 2008. Userland from a similar date is useful, as the mount(8) command needs to know about the new "log" option. With the proper system, it's pretty much a no-brainer:

• In /etc/fstab, enable logging for the file system(s) you need, in my case it's just /:      /dev/wd0a       /       ffs     rw,log        1 1 This is actually the only thing that needs to be done. All the rest writen here just explains things in a bit more details.
• Note that journaling is not active on the file system(s) at this point, so pressing the reset button for testing will result in a file system check (fsck) - don't do it right now. :)
• Reboot the system. Nothing special will show up in the boot messages:      ...     audio2 at pad0: half duplex     boot device: wd0     root on wd0a dumps on wd0b     root file system type: ffs     Fri Aug 22 20:45:55 CEST 2008     swapctl: adding /dev/wd0b as swap device at priority 0     Starting file system checks:     /dev/rwd0a: file system is clean; not checking     Setting tty flags.     ...  
• Let's recall what happens here: after probing the hardware and initializing device drivers (audio, ...), the kernel looks at disk drives for a file system with a root partition (i.e. a disk with BSD disklabel, "a" partition, and a known file system in it). It will use the first root file system it finds, and mount it read-only. As the above output is from a multi-user boot (not a single-user boot), the kernel continues to run init(8), which in turn runs /etc/rc (which then runs all of /etc/rc.d/* etc.). First things in the boot process can be determined by using the rcorder(8) tool just like /etc/rc does:
       $ cd /etc/rc.d/     $ rcorder * | head     wdogctl     raidframe     cgd     ccd     swap1     fsck     root     ... Of the above scripts, raidframe, cgd and ccd configure additional disk devices, wdogctl and swap1 are of minor interest here. The two interesting scripts are "fsck" and "root": "fsck" runs fsck(8), which in turn goes through the list of known file systems in /etc/fstab, and checks for each file system if it was unmounted cleanly last time. If not, the file system will be checked, possibly repaired, and marked as clean. This is the much-hated, time consuming process preventing a fast reboot when the system crashed.
  After ensuring all file systems are in a consistent state, the "root" script mounts the root (/) file system read-write.
  Following that, all other scripts run, create temporary files, configure network devices, enable login and whatnot. Important parts here are the order of the kernel first mounting the root file system read-only, and after checking enable writing.
• As we have marked the root file system for journaling, the log (journal) is created when mounting the file system read-write. For NetBSD, the log has only meta-data, i.e. information on what changes were made to the file system's management data structures like directories, link counts, etc. No data blocks are journaled. This may not be 100% optimal from a user point, but it ensures that the file system is in a consistent state with respect to meta-data.
• When the file system is mounted with journaling enabled, bad things are welcome (well, sort of :-) to happen, and the system will handle them gracefully: kernel panics, power failures, someone pressing the reset button - everything that disrupts system operation and gets the file system into an inconsistent state will be caught by replaying the journal on the next boot. Note that journaling will not help about user/admin errors like when you accidentally remove a file!
• After the system went down in flames -- for research purpose and better predictability, let's assume we've pressed the reset button -- with the file system in an unclean state, this will be displayed on the next boot:      ...     audio2 at pad0: half duplex     boot device: wd0     root on wd0a dumps on wd0b     /: replaying log to memory     root file system type: ffs     Fri Aug 22 20:49:55 CEST 2008     swapctl: adding /dev/wd0b as swap device at priority 0     Starting file system checks:     /dev/rwd0a: file system is journaled; not checking     /: replaying log to disk     Setting tty flags.     ...  
• After finding the root file system, the kernel first recognizes the journal, and assumes that the system crashed. The system doesn't know what's up with the disk so far, so won't go and alter the disk by writing the changes from the log onto the disk. Instead, those changes are replayed to memory only. This leaves the disk as-is, but the in-memory view of the file system will be consistent. Running fsck then recognizes the file system as journaled, and won't touch it, assuming that the log caught all bads. Mounting the file system in the next step finally replays the changes in the journal onto the disk, and finally sets it into a consistent state permanently. After that, the regular boot process can proceed as usual.
  Please note that the messages "/: replaying log to memory/disk" are printed by the kernel, as it's the kernel that runs all the file system code.
• When the system is up and running, the mount(8) command can be used to determine if logging is enabled or not:      # mount     /dev/wd0a on / type ffs (log, local) The "log" here in the mount options indicates that journaling is enabled.

First impressions of journaling are pretty good, the facts that the journal needs no further maintenance. The fact that it's placed inside the file system per default and doesn't need extra space is very nice, too. People that want to keep the log after a partition for a reason can do so, plus also specify a maximum journal size. The enduser impact of this is that lenghty file system checks are (hopefully :-) a thing of the past now!

gvim

值得试试,虽然都完成类似的功能.

cnhnln

单从他结果来看日志比soft deps更胜一筹，已经接近异步的性能了

[ 本帖最后由 cnhnln 于 2008-8-26 21:09 编辑 ]

cnhnln

Jibbed LiveCD V0.15.72 released


Zafer Aydogan has released Version 0.15.72 of his Jibbed NetBSD LiveCD. News are that the it's a CD again, instead of a DVD. Kernel sources are from NetBSD-current with kernel version 4.99.72 (as or August 19th), the CD also has Firefox 3, Roy Marples' dhcpcd 4.0rc3, WAPBL (journaling for FFS), top 3.8beta1, OpenSSH with HPN-Patch, Postfix 2.5.2, Bind 9, and may more.
Right now the CD is a bit slow as compressed VND images are currently broken - anyone who can fix this is more than welcome to do so.

有兴趣的可以试试这个livecd，带了日志文件系统