Open Stack

What is your load balancer hardware config?  Is the load balancer ingress traffic handled with just one cpu core or more?  Please run htop or top (and  press '1') to make sure all traffic isn't handled by a single haproxy process running on a single core.

If you are running haproxy 1.4, please upgrade 1.5 which brings in multiprocess+process-to-core binding support.  Please make sure you are spreading the ingress load across CPUs available as described at http://blog.onefellow.com/post/82478335338/haproxy-mapping-process-to-cpu-core-for-maximum.

Good luck.

On 8 May 2015, at 04:40, Tarak Mukherjee <tmukharj at cisco.com<mailto:tmukharj at cisco.com>> wrote:

Yes, thats correct ..
If we upload directly to swift proxy then it's indeed boost up the speed significantly .

So , haproxy ? causing the issue ..

Best Rgds,
Tarak Mukherjee

On Friday 08 May 2015 04:54 PM, Christian Schwede wrote:

Hello,

On 08.05.15 12:15, Tarak Mukherjee wrote:


We have an use-case where we upload large object (2GB - 4GB) via form
post. What we are observing is , it's taking 2 - 3hrs to get the
upload completed .

Our Setup is consist of Haproxy => Swift-Proxy => Storage Node .


any chance to test an upload without the haproxy, just uploading to the
swift proxy itself? Also, is there a HTTP proxy somewhere involved, that
the client is maybe using?

In the past I saw some real slow uploads too, but that was always
because either a http proxy was involved or some other network devices
tried to cache data (loadbalancers, proxies, ...).

-- Christian



_______________________________________________
Mailing list: http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
Post to     : openstack at lists.openstack.org<mailto:openstack at lists.openstack.org>
Unsubscribe : http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack