[Gluster-users] Fwd:Re: client is terrible with large amount of small files

[Vijay Bellur]

On 05/05/2015 03:43 PM, Kamal wrote:
> Hi Vijay,
>
>        We Tried the same. But its doesn't improve speed.
>
> For testing glusterfs, we are running storage in virtual machines. Is
> that will make any difference ?

Performance testing on physical machines is a better bet as it takes 
some variables away from the equation.

Since you happen to use a replicated volume, can you try by enabling 
this option?

gluster volume set <volname> cluster.ensure-durability off

Additionally you might want to try "strace -Tcf git clone .." and 
"gluster volume profile ..." to figure out where the latency is stemming 
from.


> But, copying the same folder between two storage machines is really
> fast. fyi.
>

Does this copying involve gluster or not?

Regards,
Vijay

> Regards,
> Kamal
>
>
>
> ---- On Tue, 05 May 2015 15:08:27 +0530 *Vijay
> Bellur<vbellur at redhat.com>* wrote ----
>
>     On 05/05/2015 12:59 PM, Kamal wrote:
>      > Hi Amukher,
>      >
>      > Even after upgrade to 3.7 small files transfer rate is slow.
>      >
>      > Below is the volume info.
>      >
>      > Volume Name: integvol1
>      > Type: Replicate
>      > Volume ID: 31793ba4-eeca-462a-a0cd-9adfb281225b
>      > Status: Started
>      > Number of Bricks: 1 x 2 = 2
>      > Transport-type: tcp
>      > Bricks:
>      > Brick1: integ-gluster1:/srv/sdb2/brick4
>      > Brick2: integ-gluster2:/srv/sdb2/brick4
>      > Options Reconfigured:
>      > server.event-threads: 30
>      > client.event-threads: 30
>      > ----
>      >
>      > I understand that for replication it would take some more time, but
>      > here its taking more time.
>      >
>      >
>      > Time taken for git clone in non gluster directory = 25 sec
>      >
>      > Time taken for git clone in gluster directory = 14 minutes
>      >
>      > Its a huge difference. Plz let me know any other tuning
>     parameters need
>      > to be done.
>      >
>      >
>
>     I have seen this before and it primarily seems to be related to the
>     readdir calls done by git clone.
>
>     Turning on these options might help to some extent:
>
>     gluster volume set <volname> performance.readdir-ahead on
>
>     gluster volume set <volname> cluster.readdir-optimize on
>
>     Please do let us know what you observe with these options enabled.
>
>     Regards,
>     Vijay
>
>
>