Hi Shrinand, Thanks for response, I understand the SWIFT limitations. Every request includes 1 put get and delete cycle so no storage on disk. I have tried same test using curl to swift and have not seen any performance degradation though there were constant fluctuation in TPS but avg TPS for any given time slot was constant. Thanks sumit
On Mon, Mar 31, 2014 at 3:26 PM, Shrinand Javadekar <[email protected] > wrote: > Hey Sumit, > > Have you tried running a similar experiment *without* using jclouds? > Swift is known to have performance bottlenecks especially when you're > trying to write to a single container. Swift uses a sqlite DB for > keeping information of the objects in a container. As you write more > objects, this DB grows in size and can start slowing things down. > > -Shri > > On Sun, Mar 30, 2014 at 10:48 PM, Sumit Gaur <[email protected]> wrote: > > Hi > > I am running Performance test (48hours) with jclouds integrated with > > Blobstore for SWIFT APIs. I choose to have default properties > > for ContextBuilder.newBuilder(). > > > > 1) I am seeing that jclouds start degrading after 20 hours of run. Though > > it is not very steep but still there is gratual decrease in TPS. > > 2) Load point of view there are 10 parallel threads hitting jclouds for 1 > > (PUT+GET+DEL) cycle. > > 3) I need to understand jclouds tunables if any of them could help in > > handling the load for consistent performance. > > > > > https://github.com/jclouds/jclouds/blob/master/core/src/main/java/org/jclouds/Constants.java > > > > > > Thanks > > sumit >
