Hi Kaleb, Thank you very much for the replay. What you have described was also my initial understanding, but in order to confirm it I made a test where using Gluster 3.4.0 on a CentOS 6.4 64bit servers, I created a replicated volume: gluster volume create 100K replica 2 192.168.1.3:/export/100K 192.168.1.4: /export/100K which I mounted under the document root on dedicated Apache server. Created 10 000 x 100K files: for i in $(seq 10000); do tr -dc "[:alpha:]" < /dev/urandom | head -c 102400 > /var/www/html/100k/f$i ; done and start querying them form a dedicated workstation using httperf in a high rate: httperf --hog --server 192.168.1.2 --wlog y,/var/tmp/100K_urls --num-calls 1 --num-conns 30000 --rate 100 I used iptraf on the gluster servers and gluster client, and the results showed that the client box is getting very close to double the total size of the files being requested (once from each gluster server) To double check, I used also the stats given by "gluster volume status Volname client" and the difference between the BytesWritten for the client, respectively for each brick, before and after the httperf execution match closely the size of the requested files (thus close to double the total size of the requested files) Do you have any suggestion why am I observing this or what supposedly is wrong with my setup or conclusions. Thank you! Kal On Wed, Sep 11, 2013 at 10:19 AM, Kaleb S. KEITHLEY <kkeithle at redhat.com>wrote: > On 09/11/2013 10:09 AM, Kal Black wrote: > >> Hello, >> I asked this question some days ago but did not get any answer so far so >> I am posting it again. >> >> Can someone explain how Gluster native client reads from a Gluster >> replicated volume? >> >> 1. Is there a mechanism that would spread requests, form a single >> client, among all bricks/servers, taking part in a replica volume? >> > > A gluster native client sends read requests to every replica in the volume. > > It reads from the first server to respond. > > Other responses are discarded. The client only reads from a single server. > > > >> 2. Would the native client take a decision where (brick/server) to fetch >> a file from based on some criteria or it will be bound to a give >> brick/server all the time? >> >> 3. Is a file being downloaded from just one of the gluster servers? >> >> I did some tests, trying to measure network traffic between gluster >> servers and a client using "iptraf" and "gluster volume status Volname >> client", which gave confusing results so I would like someone to clear >> this out for me if possible. >> >> Thank you! >> Kal >> >> >> ______________________________**_________________ >> Gluster-users mailing list >> Gluster-users at gluster.org >> http://supercolony.gluster.**org/mailman/listinfo/gluster-**users<http://supercolony.gluster.org/mailman/listinfo/gluster-users> >> >> > -- > > Kaleb > -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20130911/924bfbfc/attachment.html>
