[Gluster-users] Gluster 3.5.2 gluster, how does cache work?

[Juan José Pavlik Salles]

Hi Ben, I also thought of that, but look at this:

[root at gluster-client-1 gluster_vol]# echo 3 > /proc/sys/vm/drop_caches; dd
if=/mnt/gluster_vol/128 of=/dev/null bs=1M;
128+0 records in
128+0 records out
134217728 bytes (134 MB) copied, 6,70874 s, 20,0 MB/s
[root at gluster-client-1 gluster_vol]# echo 3 > /proc/sys/vm/drop_caches; dd
if=/mnt/gluster_vol/128 of=/dev/null bs=1M;
128+0 records in
128+0 records out
134217728 bytes (134 MB) copied, 0,160485 s, 836 MB/s
[root at gluster-client-1 gluster_vol]# echo 3 > /proc/sys/vm/drop_caches; dd
if=/mnt/gluster_vol/128 of=/dev/null bs=1M;
128+0 records in
128+0 records out
134217728 bytes (134 MB) copied, 0,148219 s, 906 MB/s
[root at gluster-client-1 gluster_vol]#

even dropping kernel caches I get a really fast answer. I think that this
is io-cache translator working on the client side for sure.


2014-08-22 12:27 GMT-03:00 Ben Turner <bturner at redhat.com>:

> ----- Original Message -----
> > From: "Juan José Pavlik Salles" <jjpavlik at gmail.com>
> > To: gluster-users at gluster.org
> > Sent: Thursday, August 21, 2014 4:07:28 PM
> > Subject: [Gluster-users] Gluster 3.5.2 gluster, how does cache work?
> >
> > Hi guys, I've been reading a bit about caching in gluster volumes, but I
> > still don't get a few things. I set up a gluster replica 2 volume like
> this:
> >
> > [root at gluster-test-1 ~]# gluster vol info vol_rep
> > Volume Name: vol_rep
> > Type: Replicate
> > Volume ID: b77db06d-2686-46c7-951f-e43bde21d8ec
> > Status: Started
> > Number of Bricks: 1 x 2 = 2
> > Transport-type: tcp
> > Bricks:
> > Brick1: gluster-test-1:/ladrillos/l1/l
> > Brick2: gluster-test-2:/ladrillos/l1/l
> > Options Reconfigured:
> > performance.cache-min-file-size: 90MB
> > performance.cache-max-file-size: 256MB
> > performance.cache-refresh-timeout: 60
> > performance.cache-size: 256MB
> > [root at gluster-test-1 ~]#
> >
> > Then I mounted the volume with gluster client on another machine. I
> created
> > an 80Mbytes file called 80, and here you have the reading test:
> >
> > [root at gluster-client-1 gluster_vol]# dd if=/mnt/gluster_vol/80
> of=/dev/null
> > bs=1M
> > 80+0 records in
> > 80+0 records out
> > 83886080 bytes (84 MB) copied, 1,34145 s, 62,5 MB/s
> > [root at gluster-client-1 gluster_vol]# dd if=/mnt/gluster_vol/80
> of=/dev/null
> > bs=1M
> > 80+0 records in
> > 80+0 records out
> > 83886080 bytes (84 MB) copied, 0,0246918 s, 3,4 GB/s
> > [root at gluster-client-1 gluster_vol]# dd if=/mnt/gluster_vol/80
> of=/dev/null
> > bs=1M
> > 80+0 records in
> > 80+0 records out
> > 83886080 bytes (84 MB) copied, 0,0195678 s, 4,3 GB/s
> > [root at gluster-client-1 gluster_vol]#
>
> You are seeing the effect of client side kernel caching.  If you want to
> see the actual throughput for reads run:
>
> sync; echo 3 > /proc/sys/vm/drop_caches; dd blah
>
> Kernel caching happens on both the client and server side, when I want to
> see uncached performance I drop caches on both clients and servers:
>
> run_drop_cache()
> {
>     for host in $MASTERNODE $NODE $CLIENT
>     do
>         ssh -i /root/.ssh/my_id root@${host} echo "Dropping cache on
> $host"
>         ssh -i /root/.ssh/my_id root@${host} sync
>         ssh -i /root/.ssh/my_id root@${host} "echo 3 >
> /proc/sys/vm/drop_caches"
>     done
> }
>
> HTH!
>
> -b
>
> > Cache is working flawlessly, (even though that 80 Mbytes is smaller than
> the
> > min-file-size value, but I don't care about it right now) what I don't
> get
> > is where cache is being stored. Is it stored on the client side or on the
> > server side? According to documentation, the io-cache translator could be
> > loaded on both sides (client and server), how can I know where it is
> being
> > loeaded? It looks like as it was being stored locally because of the
> speed,
> > but I'd like to be sure.
> >
> > Thanks!
> >
> > --
> > Pavlik Salles Juan José
> > Blog - http://viviendolared.blogspot.com
> >
> > _______________________________________________
> > Gluster-users mailing list
> > Gluster-users at gluster.org
> > http://supercolony.gluster.org/mailman/listinfo/gluster-users
>



-- 
Pavlik Salles Juan José
Blog - http://viviendolared.blogspot.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20140822/7dbfa12e/attachment.html>