Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > linux.kernel > #1360274 > unrolled thread

Re: [PATCH] zram: export the number of available comp streams

Started byMinchan Kim <minchan@kernel.org>
First post2016-03-18 01:40 +0100
Last post2016-03-23 09:00 +0100
Articles 6 — 2 participants

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.


Contents

  Re: [PATCH] zram: export the number of available comp streams Minchan Kim <minchan@kernel.org> - 2016-03-18 01:40 +0100
    Re: [PATCH] zram: export the number of available comp streams Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com> - 2016-03-18 02:10 +0100
      Re: [PATCH] zram: export the number of available comp streams Minchan Kim <minchan@kernel.org> - 2016-03-18 02:30 +0100
        Re: [PATCH] zram: export the number of available comp streams Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com> - 2016-03-21 09:00 +0100
          Re: [PATCH] zram: export the number of available comp streams Minchan Kim <minchan@kernel.org> - 2016-03-22 01:40 +0100
            Re: [PATCH] zram: export the number of available comp streams Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com> - 2016-03-23 09:00 +0100

#1360274 — Re: [PATCH] zram: export the number of available comp streams

FromMinchan Kim <minchan@kernel.org>
Date2016-03-18 01:40 +0100
SubjectRe: [PATCH] zram: export the number of available comp streams
Message-ID<rdQCm-6sf-7@gated-at.bofh.it>
Hi Sergey,

I forgot this patch until now. Sorry about that.

On Mon, Feb 01, 2016 at 10:02:48AM +0900, Sergey Senozhatsky wrote:
> Hello Minchan,
> 
> On (01/29/16 16:28), Minchan Kim wrote:
> > Hello Sergey,
> > 
> > Sorry to late response. Thesedays, I'm really busy with personal
> > stuff.
> 
> sure, no worries :)
> 
> > On Tue, Jan 26, 2016 at 09:03:59PM +0900, Sergey Senozhatsky wrote:
> > > I've been asked several very simple questions:
> > > a) How can I ensure that zram uses (or used) several compression
> > >    streams?
> > 
> > Why does he want to ensure several compression streams?
> > As you know well, zram handle it dynamically.
> > 
> > If zram cannot allocate more streams, it means the system is
> > heavily fragmented or memory pressure at that time so there
> > is no worth to add more stream, I think.
> > 
> > Could you elaborate it more why he want to know it and what
> > he expect from that?
> 
> good questions. I believe mostly it's about fine-tuning on a
> per-device basis, which is getting especially tricky when zram
> devices are used as a sort of in-memory tmp storage for various
> applications (black boxen).
> 
> > > b) What is the current number of comp streams (how much memory
> > >    does zram *actually* use for compression streams, if there are
> > >    more than one stream)?
> > 
> > Hmm, in the kernel, there are lots of example subsystem
> > we cannot know exact memory usage. Why does the user want
> > to know exact memory usage of zram? What is his concern?
> 
> certainly true. probably some of those sub-systems/drivers have some
> sort of LRU, or shrinker callbacks, to release unneeded memory back.
> zram only allocates streams, and it basically hard to tell how many:
> up to max_comp_streams, which can be larger than the number of cpus
> on the system; because we keep preemption enabled (I didn't realize
> that until I played with the patch) around
> zcomp_strm_find()/zcomp_strm_release():
> 
> 	zram_bvec_write()
> 	{
> 		...
> 		zstrm = zcomp_strm_find(zram->comp);
> >> can preempt
> 		user_mem = kmap_atomic(page);
> >> now atomic
> 		zcomp_compress()
> 		...
> 		kunmap_atomic()
> >> can preempt
> 		zcomp_strm_release()
> 		...
> 	}
> 
> so how many streams I can have on my old 4-cpus x86_64 box?
> 
> 10?
> yes.
> 
> # cat /sys/block/zram0/mm_stat
> 630484992  9288707 13103104        0 13103104    16240        0       10
> 
> 16?
> yes.
> 
> # cat /sys/block/zram0/mm_stat
> 1893117952 25296718 31354880        0 31354880    15342        0       16
> 
> 21?
> yes.
> 
> # cat /sys/block/zram0/mm_stat
> 1893167104 28499936 46616576        0 46616576    15330        0       21
> 
> do I need 21? may be no. do I nede 18? if 18 streams are needed only 10%
> of the time (I can figure it out by doing repetitive cat zramX/mm_stat),
> then I can set max_comp_streams to make 90% of applications happy, e.g.
> max_comp_streams to 10, and save some memory.
> 

Okay. Let's go back to zcomp design decade. As you remember, the reason
we separated single and multi stream code was performance caused by
locking scheme(ie, mutex_lock in single stream model was really fast
than sleep/wakeup model in multi stream).
If we could overcome that problem back then, we should have gone to
multi stream code default.

How about using *per-cpu* streams?

I remember you wanted to create max number of comp streams statically
although I didn't want at that time but I change my decision.

Let's allocate comp stream statically but remove max_comp_streams
knob. Instead, by default, zram alloctes number of streams according
to the number of online CPU.

So I think we can solve locking scheme issue in single stream
, guarantee parallel level as well as enhancing performance with
no locking.

Downside with the approach is that unnecessary memory space reserve
although zram might be used 1% of running system time. But we
should give it up for other benefits(ie, simple code, removing
max_comp_streams knob, no need to this your stat, guarantee parallel
level, guarantee consumed memory space).

What do you think about it?

[toc] | [next] | [standalone]


#1360289

FromSergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Date2016-03-18 02:10 +0100
Message-ID<rdR5o-6Vq-11@gated-at.bofh.it>
In reply to#1360274
Hello Minchan,

On (03/18/16 09:32), Minchan Kim wrote:
[..]
> > do I need 21? may be no. do I nede 18? if 18 streams are needed only 10%
> > of the time (I can figure it out by doing repetitive cat zramX/mm_stat),
> > then I can set max_comp_streams to make 90% of applications happy, e.g.
> > max_comp_streams to 10, and save some memory.
> > 
> 
> Okay. Let's go back to zcomp design decade. As you remember, the reason
> we separated single and multi stream code was performance caused by
> locking scheme(ie, mutex_lock in single stream model was really fast
> than sleep/wakeup model in multi stream).
> If we could overcome that problem back then, we should have gone to
> multi stream code default.

yes, IIRC I wanted to limit the number of streams by the number of
online CPUs (or was it 2*num_online_cpus()?), and thus change the
number of streams dynamically (because CPUs can go on and off line);
and create at least num_online_cpus() streams during device
initialization.

the reason for a single-stream zram IIRC were setups in which
zram is used as a swap device. streams require some memory, after
all. and then we discovered that mutex spin on owner boosts single
stream zram significantly.

> How about using *per-cpu* streams?

OK. instead of list of idle streams use per-cpu pointer and process
CPU_FOO notifications. that can work, sounds good to me.


> I remember you wanted to create max number of comp streams statically
> although I didn't want at that time but I change my decision.
> 
> Let's allocate comp stream statically but remove max_comp_streams
> knob. Instead, by default, zram alloctes number of streams according
> to the number of online CPU.

OK. removing `max_comp_streams' will take another 2 years. That's
a major change, we can leave it for longer, just make it nop.

> So I think we can solve locking scheme issue in single stream
> , guarantee parallel level as well as enhancing performance with
> no locking.
>
> Downside with the approach is that unnecessary memory space reserve
> although zram might be used 1% of running system time. But we
> should give it up for other benefits

aha, ok.

> (ie, simple code, removing
> max_comp_streams knob, no need to this your stat, guarantee parallel
> level, guarantee consumed memory space).

I'll take a look and prepare some numbers (most likely next week).


> What do you think about it?

so should I ask Andrew to drop this patch?

	-ss

[toc] | [prev] | [next] | [standalone]


#1360297

FromMinchan Kim <minchan@kernel.org>
Date2016-03-18 02:30 +0100
Message-ID<rdRoK-72X-3@gated-at.bofh.it>
In reply to#1360289
On Fri, Mar 18, 2016 at 10:09:37AM +0900, Sergey Senozhatsky wrote:
> Hello Minchan,
> 
> On (03/18/16 09:32), Minchan Kim wrote:
> [..]
> > > do I need 21? may be no. do I nede 18? if 18 streams are needed only 10%
> > > of the time (I can figure it out by doing repetitive cat zramX/mm_stat),
> > > then I can set max_comp_streams to make 90% of applications happy, e.g.
> > > max_comp_streams to 10, and save some memory.
> > > 
> > 
> > Okay. Let's go back to zcomp design decade. As you remember, the reason
> > we separated single and multi stream code was performance caused by
> > locking scheme(ie, mutex_lock in single stream model was really fast
> > than sleep/wakeup model in multi stream).
> > If we could overcome that problem back then, we should have gone to
> > multi stream code default.
> 
> yes, IIRC I wanted to limit the number of streams by the number of
> online CPUs (or was it 2*num_online_cpus()?), and thus change the
> number of streams dynamically (because CPUs can go on and off line);
> and create at least num_online_cpus() streams during device
> initialization.
> 
> the reason for a single-stream zram IIRC were setups in which
> zram is used as a swap device. streams require some memory, after
> all. and then we discovered that mutex spin on owner boosts single
> stream zram significantly.
> 
> > How about using *per-cpu* streams?
> 
> OK. instead of list of idle streams use per-cpu pointer and process
> CPU_FOO notifications. that can work, sounds good to me.
> 
> 
> > I remember you wanted to create max number of comp streams statically
> > although I didn't want at that time but I change my decision.
> > 
> > Let's allocate comp stream statically but remove max_comp_streams
> > knob. Instead, by default, zram alloctes number of streams according
> > to the number of online CPU.
> 
> OK. removing `max_comp_streams' will take another 2 years. That's
> a major change, we can leave it for longer, just make it nop.
> 
> > So I think we can solve locking scheme issue in single stream
> > , guarantee parallel level as well as enhancing performance with
> > no locking.
> >
> > Downside with the approach is that unnecessary memory space reserve
> > although zram might be used 1% of running system time. But we
> > should give it up for other benefits
> 
> aha, ok.
> 
> > (ie, simple code, removing
> > max_comp_streams knob, no need to this your stat, guarantee parallel
> > level, guarantee consumed memory space).
> 
> I'll take a look and prepare some numbers (most likely next week).

Sounds great to me!

> 
> 
> > What do you think about it?
> 
> so should I ask Andrew to drop this patch?

Yeb.

Thanks!

> 
> 	-ss

[toc] | [prev] | [next] | [standalone]


#1361565

FromSergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Date2016-03-21 09:00 +0100
Message-ID<rf2UN-2uv-3@gated-at.bofh.it>
In reply to#1360297
Hello Minchan,

On (03/18/16 10:25), Minchan Kim wrote:
[..]
> > aha, ok.
> > 
> > > (ie, simple code, removing
> > > max_comp_streams knob, no need to this your stat, guarantee parallel
> > > level, guarantee consumed memory space).
> > 
> > I'll take a look and prepare some numbers (most likely next week).
> 
> Sounds great to me!

so I have schematically this thing now. streams are per-cpu and contain
scratch buffer and work mem.

zram_bvec_write()
{
	*get_cpu_ptr(comp->stream);
	 zcomp_compress();
	 zs_malloc()
	put_cpu_ptr(comp->stream);
}

this, however, makes zsmalloc unhapy. pool has GFP_NOIO | __GFP_HIGHMEM
gfp, and GFP_NOIO is ___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM. this
__GFP_DIRECT_RECLAIM is in the conflict with per-cpu streams, because
per-cpu streams require disabled preemption (up until we copy stream
buffer to zspage). so what options do we have here... from the top of
my head (w/o a lot of thinking)...
-- remove __GFP_DIRECT_RECLAIM from pool gfp mask, which a bit is risky...
IOW, make pool gfp '___GFP_KSWAPD_RECLAIM | __GFP_HIGHMEM'
-- kmalloc/kfree temp buffer for every RW op, which is ugly... because
it sort of voids the whole purpose of per-cpu streams.
-- ... hm.

	-ss

[toc] | [prev] | [next] | [standalone]


#1362244

FromMinchan Kim <minchan@kernel.org>
Date2016-03-22 01:40 +0100
Message-ID<rfiwy-4XQ-11@gated-at.bofh.it>
In reply to#1361565
Hello Sergey,

On Mon, Mar 21, 2016 at 04:51:28PM +0900, Sergey Senozhatsky wrote:
> Hello Minchan,
> 
> On (03/18/16 10:25), Minchan Kim wrote:
> [..]
> > > aha, ok.
> > > 
> > > > (ie, simple code, removing
> > > > max_comp_streams knob, no need to this your stat, guarantee parallel
> > > > level, guarantee consumed memory space).
> > > 
> > > I'll take a look and prepare some numbers (most likely next week).
> > 
> > Sounds great to me!
> 
> so I have schematically this thing now. streams are per-cpu and contain
> scratch buffer and work mem.
> 
> zram_bvec_write()
> {
> 	*get_cpu_ptr(comp->stream);
> 	 zcomp_compress();
> 	 zs_malloc()
> 	put_cpu_ptr(comp->stream);
> }
> 
> this, however, makes zsmalloc unhapy. pool has GFP_NOIO | __GFP_HIGHMEM
> gfp, and GFP_NOIO is ___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM. this
> __GFP_DIRECT_RECLAIM is in the conflict with per-cpu streams, because
> per-cpu streams require disabled preemption (up until we copy stream
> buffer to zspage). so what options do we have here... from the top of
> my head (w/o a lot of thinking)...

Indeed.

> -- remove __GFP_DIRECT_RECLAIM from pool gfp mask, which a bit is risky...
> IOW, make pool gfp '___GFP_KSWAPD_RECLAIM | __GFP_HIGHMEM'

Yeb. It would be okay for zram-swap but not zram-blk.

> -- kmalloc/kfree temp buffer for every RW op, which is ugly... because
> it sort of voids the whole purpose of per-cpu streams.

How about this?

zram_bvec_write()
{
retry:
        *get_cpu_ptr(comp->stream);
        zcomp_compress();
        handle = zs_malloc((gfp &~ __GFP_DIRECT_RECLAIM| | GFP_NOWARN)
        if (!handle) {
                put_cpu_ptr(comp->stream);
                handle  = zs_malloc(gfp);
                goto retry;
        }
        put_cpu_ptr(comp->stream);
}

If per-cpu model really performance win, it is worth to try.

[toc] | [prev] | [next] | [standalone]


#1363232

FromSergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Date2016-03-23 09:00 +0100
Message-ID<rfLRV-g1-17@gated-at.bofh.it>
In reply to#1362244
( was "[PATCH] zram: export the number of available comp streams"
  forked from http://marc.info/?l=linux-kernel&m=145860707516861 )


Hello Minchan,

forked into a separate tread.

On (03/22/16 09:39), Minchan Kim wrote:
> > zram_bvec_write()
> > {
> > 	*get_cpu_ptr(comp->stream);
> > 	 zcomp_compress();
> > 	 zs_malloc()
> > 	put_cpu_ptr(comp->stream);
> > }
> > 
> > this, however, makes zsmalloc unhapy. pool has GFP_NOIO | __GFP_HIGHMEM
> > gfp, and GFP_NOIO is ___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM. this
> > __GFP_DIRECT_RECLAIM is in the conflict with per-cpu streams, because
> > per-cpu streams require disabled preemption (up until we copy stream
> > buffer to zspage). so what options do we have here... from the top of
> > my head (w/o a lot of thinking)...
> 
> Indeed.
...
> How about this?
> 
> zram_bvec_write()
> {
> retry:
>         *get_cpu_ptr(comp->stream);
>         zcomp_compress();
>         handle = zs_malloc((gfp &~ __GFP_DIRECT_RECLAIM| | GFP_NOWARN)
>         if (!handle) {
>                 put_cpu_ptr(comp->stream);
>                 handle  = zs_malloc(gfp);
>                 goto retry;
>         }
>         put_cpu_ptr(comp->stream);
> }

interesting. the retry jump should go higher, we have "user_mem = kmap_atomic(page)"
which we unmap right after compression, because a) we don't need
uncompressed memory anymore b) zs_malloc() can sleep and we can't have atomic
mapping around. the nasty thing here is is_partial_io(). we need to re-do

	if (is_partial_io(bvec))
		memcpy(uncmem + offset, user_mem + bvec->bv_offset,
			bvec->bv_len);

once again in the worst case.

so zs_malloc((gfp &~ __GFP_DIRECT_RECLAIM | GFP_NOWARN) so far can cause
double memcpy() and double compression. just to outline this.


the test.

I executed a number of iozone tests, on each iteration re-creating zram
device (3GB, LZO, EXT4. the box has 4 x86_64 CPUs).

$DEVICE_SZ=3G
$FREE_SPACE is 10% of $DEVICE_SZ
time ./iozone -t $i -R -r $((8*$i))K -s $((($DEVICE_SZ/$i - $FREE_SPACE)/(1024*1024)))M -I +Z


columns:

       TEST           MAX_STREAMS 4   MAX_STREAMS 8  PER_CPU STREAMS
====================================================================

Test #1 iozone -t 1 -R -r 8K -s 2764M -I +Z
  Initial write         853492.31*      835868.50       839789.56
        Rewrite        1642073.88      1657255.75      1693011.50*
           Read        3384044.00*     3218727.25      3269109.50
        Re-read        3389794.50*     3243187.00      3267422.25
   Reverse Read        3209805.75*     3082040.00      3107957.25
    Stride read        3100144.50*     2972280.25      2923155.25
    Random read        2992249.75*     2874605.00      2854824.25
 Mixed workload        2992274.75*     2878212.25      2883840.00
   Random write        1471800.00      1452346.50      1515678.75*
         Pwrite         802083.00       801627.31       820251.69*
          Pread        3443495.00*     3308659.25      3302089.00
         Fwrite        1880446.88      1838607.50      1909490.00*
          Fread        3479614.75      3091634.75      6442964.50*
=          real          1m4.170s        1m4.513s        1m4.123s
=          user          0m0.559s        0m0.518s        0m0.511s
=           sys         0m18.766s       0m19.264s       0m18.641s


Test #2 iozone -t 2 -R -r 16K -s 1228M -I +Z
  Initial write        2102532.12      2051809.19      2419072.50*
        Rewrite        2217024.25      2250930.00      3681559.00*
           Read        7716933.25      7898759.00      8345507.75*
        Re-read        7748487.75      7765282.25      8342367.50*
   Reverse Read        7415254.25      7552637.25      7822691.75*
    Stride read        7041909.50      7091049.25      7401273.00*
    Random read        6205044.25      6738888.50      7232104.25*
 Mixed workload        4582990.00      5271651.50      5361002.88*
   Random write        2591893.62      2513729.88      3660774.38*
         Pwrite        1873876.75      1909758.69      2087238.81*
          Pread        4669850.00      4651121.56      4919588.44*
         Fwrite        1937947.25      1940628.06      2034251.25*
          Fread        9930319.00      9970078.00*     9831422.50
=          real         0m53.844s       0m53.607s       0m52.528s
=          user          0m0.273s        0m0.289s        0m0.280s
=           sys         0m16.595s       0m16.478s       0m14.072s


Test #3 iozone -t 3 -R -r 24K -s 716M -I +Z
  Initial write        3036567.50      2998918.25      3683853.00*
        Rewrite        3402447.88      3415685.88      5054705.38*
           Read       11767413.00*    11133789.50     11246497.25
        Re-read       11797680.50*    11092592.00     11277382.00
   Reverse Read       10828320.00*    10157665.50     10749055.00
    Stride read       10532039.50*     9943521.75     10464700.25
    Random read       10380365.75*     9807859.25     10234127.00
 Mixed workload        8772132.50*     8415083.50      8457108.50
   Random write        3364875.00      3310042.00      5059136.38*
         Pwrite        2677290.25      2651309.50      3198166.25*
          Pread        5221799.56*     4963050.69      4987293.78
         Fwrite        2026887.56      2047679.00      2124199.62*
          Fread       11310381.25     11413531.50     11444208.75*
=          real         0m50.209s       0m50.782s       0m49.750s
=          user          0m0.195s        0m0.205s        0m0.215s
=           sys         0m14.873s       0m15.159s       0m12.911s


Test #4 iozone -t 4 -R -r 32K -s 460M -I +Z
  Initial write        3841474.94      3859279.81      5309988.88*
        Rewrite        3905526.25      3917309.62      6814800.62*
           Read       16233054.50     14843560.25     16352283.75*
        Re-read       16335506.50     15529152.25     16352570.00*
   Reverse Read       15316394.50*    14225482.50     15004897.50
    Stride read       14799380.25*    14064034.25     14355184.25
    Random read       14683771.00     14206928.50     14814913.00*
 Mixed workload        9058851.50      9180650.75     10815917.50*
   Random write        3990585.94      4004757.00      6722088.50*
         Pwrite        3318836.12      3468977.69      4244747.69*
          Pread        5894538.16*     5588046.38      5847345.62
         Fwrite        2227353.75      2186688.62      2386974.88*
          Fread       12046094.00     12240004.75*    12073956.75
=          real         0m48.561s       0m48.839s       0m48.142s
=          user          0m0.155s        0m0.170s        0m0.133s
=           sys         0m13.650s       0m13.684s       0m10.790s


Test #5 iozone -t 5 -R -r 40K -s 307M -I +Z
  Initial write        4034878.94      4026610.69      5775746.12*
        Rewrite        3898600.44      3901114.16      6923764.19*
           Read       14947360.88     16698824.25*    10155333.62
        Re-read       15844580.75*    15344057.00      9869874.38
   Reverse Read        7459156.95      9023317.86*     7648295.03
    Stride read       10823891.81      9615553.81     11231183.72*
    Random read       10391702.56*     9740935.75     10048038.28
 Mixed workload        8261830.94     10175925.00*     7535763.75
   Random write        3951423.31      3960984.62      6671441.38*
         Pwrite        4119023.12      4097204.56      5975659.12*
          Pread        6072076.73*     4338668.50      6020808.34
         Fwrite        2417235.47      2337875.88      2665450.62*
          Fread       13393630.25     13648332.00*    13395391.00
=          real         0m47.756s       0m47.939s       0m47.483s
=          user          0m0.128s        0m0.128s        0m0.119s
=           sys         0m10.361s       0m10.392s        0m8.717s


Test #6 iozone -t 6 -R -r 48K -s 204M -I +Z
  Initial write        4134932.97      4137171.88      5983193.31*
        Rewrite        3928131.31      3950764.00      7124248.00*
           Read       10965005.75*    10152236.50      9856572.88
        Re-read        9386946.00     10776231.38     14303174.12*
   Reverse Read        6035244.89      7456152.38*     5999446.38
    Stride read        8041000.75      7995307.75     10182936.75*
    Random read        8565099.09     10487707.58*     8694877.25
 Mixed workload        5301593.06      7332589.09*     6802251.06
   Random write        4046482.56      3986854.94      6723824.56*
         Pwrite        4188226.41      4214513.34      6245278.44*
          Pread        3452596.86      3708694.69*     3486420.41
         Fwrite        2829500.22      3030742.72      3033792.28*
          Fread       13331387.75     13490416.50     14940410.25*
=          real         0m47.150s       0m47.050s       0m47.044s
=          user          0m0.106s        0m0.100s        0m0.094s
=           sys          0m9.238s        0m8.804s        0m6.930s


Test #7 iozone -t 7 -R -r 56K -s 131M -I +Z
  Initial write        4169480.84      4116331.03      5946801.38*
        Rewrite        3993155.97      3986195.00      6928142.44*
           Read       18901600.25*    10088918.69      6699592.78
        Re-read        8738544.69     14881309.62*    13960026.06
   Reverse Read        5008919.08      7923949.95*     5495212.41
    Stride read        7029436.75      8747574.91*     6477087.25
    Random read        6994738.56*     5448687.81      6585235.53
 Mixed workload        5178632.44      5258914.92      5587421.81*
   Random write        4008977.78      3928116.88      6816453.12*
         Pwrite        4342852.09      4154319.09      6124520.06*
          Pread        3880318.99      2978587.56      4493903.14*
         Fwrite        5557990.03      2923556.59      6126649.94*
          Fread       14451722.00     15281179.62*    14675436.50
=          real         0m46.321s       0m46.458s       0m45.791s
=          user          0m0.093s        0m0.089s        0m0.095s
=           sys          0m6.961s        0m6.600s        0m5.499s


Test #8 iozone -t 8 -R -r 64K -s 76M -I +Z
  Initial write        4354783.88      4392731.31      6337397.50*
        Rewrite        4070162.69      3974051.50      7587279.81*
           Read       10095324.56     17945227.88*     8359665.56
        Re-read       12316555.88     20468303.75*     7949999.34
   Reverse Read        4924659.84      8542573.33*     6388858.72
    Stride read       10895715.69     14828968.38*     6107484.81
    Random read        6838537.34     14352104.25*     5389174.97
 Mixed workload        5805646.75      8391745.53*     6052748.25
   Random write        4148973.38      3890847.38      7247214.19*
         Pwrite        4309372.41      4423800.34      6863604.69*
          Pread        4875766.02*     4042375.33      3692948.91
         Fwrite        6102404.31      6021884.41      6634112.09*
          Fread       15485971.12*    14900780.62     13981842.50
=          real         0m45.618s       0m45.753s       0m45.619s
=          user          0m0.071s        0m0.080s        0m0.060s
=           sys          0m4.702s        0m4.430s        0m3.555s


Test #9 iozone -t 9 -R -r 72K -s 34M -I +Z
  Initial write        4202354.67      4208936.34      6300798.88*
        Rewrite        4046855.38      4294137.50      7623323.69*
           Read       10926571.88     13304801.81*    10895587.19
        Re-read       17725984.94*     7964431.25     12394078.50
   Reverse Read        5843121.72      5851846.66*     4075657.20
    Stride read        9688998.59     10306234.70*     5566376.62
    Random read        7656689.97      8660602.06*     5437182.36
 Mixed workload        6229215.62     11205238.73*     5575719.75
   Random write        4094822.22      4517401.86      6601624.94*
         Pwrite        4274497.50      4263936.64      6844453.11*
          Pread        6525075.62*     6043725.62      5745003.28
         Fwrite        5958798.56      8430354.78*     7636085.00
          Fread       18636725.12*    17268959.12     16618803.62
=          real         0m44.945s       0m44.816s       0m45.194s
=          user          0m0.062s        0m0.060s        0m0.060s
=           sys          0m2.187s        0m2.223s        0m1.888s


Test #10 iozone -t 10 -R -r 80K -s 0M -I +Z
  Initial write        3213973.56      2731512.62      4416466.25*
        Rewrite        3066956.44*     2693819.50       332671.94
           Read        7769523.25*     2681473.75       462840.44
        Re-read        5244861.75      5473037.00*      382183.03
   Reverse Read        7479397.25*     4869597.75       374714.06
    Stride read        5403282.50*     5385083.75       382473.44
    Random read        5131997.25      5176799.75*      380593.56
 Mixed workload        3998043.25      4219049.00*     1645850.45
   Random write        3452832.88      3290861.69      3588531.75*
         Pwrite        3757435.81      2711756.47      4561807.88*
          Pread        2743595.25*     2635835.00       412947.98
         Fwrite       16076549.00     16741977.25*    14797209.38
          Fread       23581812.62*    21664184.25      5064296.97
=          real         0m44.490s       0m44.444s       0m44.609s
=          user          0m0.054s        0m0.049s        0m0.055s
=           sys          0m0.037s        0m0.046s        0m0.148s


so when the number of active tasks become larger than the number
of online CPUS, iozone reports a bit hard to understand data. I
can assume that since now we keep the preemption disabled longer
in write path, a concurrent operation (READ or WRITE) cannot preempt
current anymore... slightly suspicious.

the other hard to understand thing is why do READ-only tests have
such a huge jitter. READ-only tests don't depend on streams, they
don't even use them, we supply compressed data directly to
decompression api.

may be better retire iozone and never use it again.


"118 insertions(+), 238 deletions(-)" the patches remove a big
pile of code.

	-ss

[toc] | [prev] | [standalone]


Back to top | Article view | linux.kernel


csiph-web