Re: suspiciously good fsck times?

Re: suspiciously good fsck times?

[Ric Wheeler]

Theodore Tso wrote:
> On Thu, Jul 10, 2008 at 11:14:28AM -0500, Eric Sandeen wrote:
>   
>> Val & I talked about this a little, and came to the conclusion that
>> directory fragmentation might be a pretty big part of it.
>>     
>
> Hmm, could be.  Let's see.  Ric said 46.5 million files, I don't know
> how big the filenames were, but let's assume a directory entry size of
> 32, so that means if we assume perfect packing, 128 directory entries
> per 4k block.  Let's use 100 directory entries/blok just to make the
> math easyer, so that's 465,000 blocks.  If we assume a 10ms seek time,
> and that the blocks are totally scattered, that's 4650 seconds, or
> 1.29 hours. So that's roughly within the ballpark that Ric measured.
>
>      	       	      	      	     	 	  - Ted
>   

(changing cc to the real list instead of ext4 owner - sorry!)

The file names are 40 bytes long, (6 initial bytes of time stamp with 24 
random bytes at end of name. For example:

451aeb61ead89~~~DYASX8LYL4NAUWK3WI187VRP

The 4 threads chose the target subdirectory based on the time stamp, 
rotating into a new subdirectory every 3 minutes or so.

ric

suspiciously good fsck times?

[Ric Wheeler]

(Repost to the list - this was mistakingly sent to linux-ext4-owner)

Just to be mean, I have been trying to test the fsck speed of ext4 with
lots of small files.  The test I ran uses fs_mark to fill a 1TB Seagate
drive with 45.6 million 20k files (distributed between 256 subdirectories).

Running on ext3, "fsck -f" takes about one hour.

Running on ext4, with uninit_bg, the same fsck is finished in a bit over
5 minutes - more than 10x faster.  (Without uninit_bg, the fsck takes
about 10 minutes).

Is this too good to be true? Below is the fsck run itself, the tree is
Ted's latest git tree and his 1.41 WIP tools,

ric


[root@localhost Perf]# time /sbin/fsck.ext4 -t -t -f /dev/sdb1
e4fsck 1.41-WIP (07-Jul-2008)
Pass 1: Checking inodes, blocks, and sizes
Pass 1: Memory used: 40632k/69424k (36424k/4209k), time: 204.95/78.22/25.58
Pass 1: I/O read: 11140MB, write: 0MB, rate: 54.35MB/s
Pass 2: Checking directory structure
Pass 2: Memory used: 70184k/61968k (51803k/18382k), time: 76.47/50.27/ 8.77
Pass 2: I/O read: 3023MB, write: 0MB, rate: 39.53MB/s
Pass 3: Checking directory connectivity
Peak memory: Memory used: 70184k/61968k (59256k/10929k), time:
281.72/128.59/34.35
Pass 3A: Memory used: 70184k/61968k (59256k/10929k), time:  0.00/ 0.00/
0.00
Pass 3A: I/O read: 0MB, write: 0MB, rate: 0.00MB/s
Pass 3: Memory used: 70184k/61968k (51803k/18382k), time:  0.03/ 0.00/ 0.00
Pass 3: I/O read: 1MB, write: 0MB, rate: 37.86MB/s
Pass 4: Checking reference counts
Pass 4: Memory used: 70184k/44968k (27354k/42831k), time:  2.37/ 2.36/ 0.00
Pass 4: I/O read: 0MB, write: 0MB, rate: 0.00MB/s
Pass 5: Checking group summary information
Pass 5: Memory used: 70184k/240k (64619k/5566k), time: 19.40/ 5.52/ 0.29
Pass 5: I/O read: 34MB, write: 0MB, rate: 1.75MB/s
/dev/sdb1: 45600268/61054976 files (0.0% non-contiguous),
232657574/244190000 blocks
Memory used: 70184k/240k (64889k/5296k), time: 303.54/136.48/34.65
I/O read: 14198MB, write: 1MB, rate: 46.77MB/s

real    5m3.993s
user    2m16.477s
sys     0m35.041s

Re: suspiciously good fsck times?

[Ric Wheeler]

Theodore Tso wrote:
> On Thu, Jul 10, 2008 at 08:36:42AM -0400, Ric Wheeler wrote:
>   
>> Just to be mean, I have been trying to test the fsck speed of ext4 with  
>> lots of small files.  The test I ran uses fs_mark to fill a 1TB Seagate  
>> drive with 45.6 million 20k files (distributed between 256 
>> subdirectories).
>>
>> Running on ext3, "fsck -f" takes about one hour.
>>
>> Running on ext4, with uninit_bg, the same fsck is finished in a bit over  
>> 5 minutes - more than 10x faster.  (Without uninit_bg, the fsck takes  
>> about 10 minutes).
>>
>> Is this too good to be true? Below is the fsck run itself, the tree is  
>> Ted's latest git tree and his 1.41 WIP tools,
>>     
>
> Wow.  My guess is that flex_bg is making the difference.  What we
> would want to compare is the I/O read statistics line:
>
>   
>> I/O read: 14198MB, write: 1MB, rate: 46.77MB/s
>>     
>
> That's pretty good, and indicates we've avoided a *lot* of seeking.
> The e2fsck -t -t output for ext3 should show roughly the same mount of
> I/O read (with 20k files, there would be no advantage towards using
> extents), but the I/O rate is probably *much* lower, indicating a lot
> more seeking is going on.
>   
We did run fsck through seekwatcher & saw a significant reduction in
seeks/sec for ext4. Eric has the pretty pictures that he can share.

> Can you send the full e2fsck -t -t output of the ext3 run?  And what
> is the hdparm -t -t results of the disk?
>   

I didn't run the ext3 test with -t -t (but can refill and rerun, takes
about 12 hours).

This disk is a relatively new Seagate 1TB drive, specs at:

http://www.seagate.com/ww/v/index.jsp?vgnextoid=0732f141e7f43110VgnVCM100000f5ee0a0aRCRD

hdparm test:

[root@localhost rwheeler]# /sbin/hdparm -t -t /dev/sdb

/dev/sdb:
Timing buffered disk reads:  186 MB in  3.03 seconds =  61.33 MB/sec



> If I'm right, if you create the filesystem with mke2fs -t ext4dev -O
> ^flex_bg,^uninit_bg, you should see performance back to the old ext3
> levels.
>   

With uninit_bg off, it ran about 10 minutes, but it would be interesting
to run without either.
> 							- Ted
>
> P.S.  We probably do want to examine the block allocation layout with
> flex_bg to make sure that the filesystem ages well in the long term.
>   
Testing aged file systems is always the holy grail - this workload is a
fairly artificial one and was laid down with 4 threads currently writing
to a shared subdirectory.

ric

Re: suspiciously good fsck times?

[Theodore Tso]

Based on the graphs which Eric posted, One interesting thing I think
you'll find if you repeat the ext3 experiment with e2fsck -t -t is
that pass2 will be about seven times longer than pass1.  (Which is
backwards from most e2fsck runs, where pass2 is about half pass 1's
run time --- although obviously that depends on how many directory
blocks you have.)

Yes, some kind of reservation windows would help on ext3 --- but the
question is whether such a change would be too-specific for this
benchmark or not.  Most of the time directories don't grow to such a
huge size.  So if you use a smallish (around 8 blocks, say) for many
directories this might lead to more filesystem fragmentation that in
the long run would cause the filesystem not to age well; it also
wouldn't help much when you have over 11 million files in the
directory, and a directory with over 100,000 blocks.

I don't think delayed allocation is what's helping here either,
because the journal will force the directory blocks to be placed as
soon as we commit a transaction.  I think what's saving us here is
that flex_bg and mballoc is separating the directory blocks from the
data blocks, allowng the directory blocks to be closely packed
together.

					- Ted

Re: suspiciously good fsck times?

[Ric Wheeler]

Theodore Tso wrote:
> Based on the graphs which Eric posted, One interesting thing I think
> you'll find if you repeat the ext3 experiment with e2fsck -t -t is
> that pass2 will be about seven times longer than pass1.  (Which is
> backwards from most e2fsck runs, where pass2 is about half pass 1's
> run time --- although obviously that depends on how many directory
> blocks you have.)
>
>   
Pass2 was where both spent most of their time, but I can rerun later to 
validate that.

> Yes, some kind of reservation windows would help on ext3 --- but the
> question is whether such a change would be too-specific for this
> benchmark or not.  Most of the time directories don't grow to such a
> huge size.  So if you use a smallish (around 8 blocks, say) for many
> directories this might lead to more filesystem fragmentation that in
> the long run would cause the filesystem not to age well; it also
> wouldn't help much when you have over 11 million files in the
> directory, and a directory with over 100,000 blocks.
>   
I think that the key is to lay out the directories (or files for that 
matter) in reasonably contiguous chunks.  If we could always bump up the 
allocation by enough to capture a full disk track (128k? 512k?) you 
would probably be near optimal, but any significant portion of a track 
would also help.

It would be interesting to rerun with the 46 million files in one 
directory as well (basically, for working sets that have no natural 
mapping into directories like some object based workloads).

> I don't think delayed allocation is what's helping here either,
> because the journal will force the directory blocks to be placed as
> soon as we commit a transaction.  I think what's saving us here is
> that flex_bg and mballoc is separating the directory blocks from the
> data blocks, allowng the directory blocks to be closely packed
> together.
>
> 					- Ted
>   
I can try to validate that, thanks!

ric

Re: suspiciously good fsck times?

[Ric Wheeler]

Theodore Tso wrote:
> Based on the graphs which Eric posted, One interesting thing I think
> you'll find if you repeat the ext3 experiment with e2fsck -t -t is
> that pass2 will be about seven times longer than pass1.  (Which is
> backwards from most e2fsck runs, where pass2 is about half pass 1's
> run time --- although obviously that depends on how many directory
> blocks you have.)
>
> Yes, some kind of reservation windows would help on ext3 --- but the
> question is whether such a change would be too-specific for this
> benchmark or not.  Most of the time directories don't grow to such a
> huge size.  So if you use a smallish (around 8 blocks, say) for many
> directories this might lead to more filesystem fragmentation that in
> the long run would cause the filesystem not to age well; it also
> wouldn't help much when you have over 11 million files in the
> directory, and a directory with over 100,000 blocks.
>
> I don't think delayed allocation is what's helping here either,
> because the journal will force the directory blocks to be placed as
> soon as we commit a transaction.  I think what's saving us here is
> that flex_bg and mballoc is separating the directory blocks from the
> data blocks, allowng the directory blocks to be closely packed
> together.
>
> 					- Ted
>   

I made a new ext4 file system without flex_bg or uninit:[

root@localhost Perf]# /sbin/debuge4fs /dev/sdb1
debuge4fs 1.41-WIP (07-Jul-2008)
debuge4fs:  feature
Filesystem features: has_journal ext_attr resize_inode dir_index 
filetype extent sparse_super large_file


The fsck time was a bit slower, but still looks like 8 minutes on ext4 
vs 1 hour on ext3:

[root@localhost Perf]# umount /mnt
[root@localhost Perf]# time /sbin/fsck.ext4 -t -t -f /dev/sdb1
e4fsck 1.41-WIP (07-Jul-2008)
Pass 1: Checking inodes, blocks, and sizes
Pass 1: Memory used: 43944k/69424k (36476k/7469k), time: 352.48/93.27/29.45
Pass 1: I/O read: 14914MB, write: 0MB, rate: 42.31MB/s
Pass 2: Checking directory structure
Pass 2: Memory used: 71396k/61968k (51854k/19543k), time: 73.00/50.46/ 7.65
Pass 2: I/O read: 3023MB, write: 0MB, rate: 41.41MB/s
Pass 3: Checking directory connectivity
Peak memory: Memory used: 71396k/61968k (59307k/12090k), time: 
425.82/143.83/37.10
Pass 3A: Memory used: 71396k/61968k (59307k/12090k), time:  0.00/ 0.00/ 0.00
Pass 3A: I/O read: 0MB, write: 0MB, rate: 0.00MB/s
Pass 3: Memory used: 71396k/61968k (51854k/19543k), time:  0.01/ 0.00/ 0.00
Pass 3: I/O read: 1MB, write: 0MB, rate: 76.91MB/s
Pass 4: Checking reference counts
Pass 4: Memory used: 71396k/44968k (27406k/43991k), time:  2.37/ 2.36/ 0.00
Pass 4: I/O read: 0MB, write: 0MB, rate: 0.00MB/s
Pass 5: Checking group summary information
Pass 5: Memory used: 71396k/240k (64671k/6726k), time: 63.60/ 4.98/ 0.33
Pass 5: I/O read: 37MB, write: 0MB, rate: 0.58MB/s
/dev/sdb1: 45600268/61054976 files (0.0% non-contiguous), 
232657587/244190000 blocks
Memory used: 71396k/240k (64671k/6726k), time: 491.82/151.17/37.43
I/O read: 17974MB, write: 1MB, rate: 36.55MB/s

real    8m12.260s
user    2m31.167s
sys     0m37.766s

Re: suspiciously good fsck times?

[Andreas Dilger]

On Jul 10, 2008  16:13 -0400, Ric Wheeler wrote:
> It would be interesting to rerun with the 46 million files in one  
> directory as well (basically, for working sets that have no natural  
> mapping into directories like some object based workloads).

I think you'll hit a limit around 15M files in a single directory.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.

Re: suspiciously good fsck times?

[Ric Wheeler]

Andreas Dilger wrote:
> On Jul 10, 2008  16:13 -0400, Ric Wheeler wrote:
>   
>> It would be interesting to rerun with the 46 million files in one  
>> directory as well (basically, for working sets that have no natural  
>> mapping into directories like some object based workloads).
>>     
>
> I think you'll hit a limit around 15M files in a single directory.
>
> Cheers, Andreas
> --
>   
Probably still worth a quick test, just to see how well it holds up at 
the edge, thanks!

ric