Can I generalise that 5100 ops / elapsed time or 5100 / 1.499 or 3402 records/s ?
On Saturday, February 1, 2014 11:57:35 PM UTC, asdf9898 wrote: > > Hi, > > > I'm trying to understand the RoadRunner output and translate into simple > records per second throughput > > So for 100 records (-d 100) that would mean 100 inserts and 100 reads = > 200 operations > > In the output below what does the 5100 ops mean? > > > 23:52:32.591 [main] INFO com.couchbase.roadrunner.RoadRunner - > Operations: measured 5100ops out of total 5100ops. > 23:52:32.600 [main] INFO com.couchbase.roadrunner.RoadRunner - Percentile > (microseconds) for "get" Workload: > 23:52:32.601 [main] INFO com.couchbase.roadrunner.RoadRunner - 50%:278 > 75%:318 95%:365 99%:479 > 23:52:32.601 [main] INFO com.couchbase.roadrunner.RoadRunner - Percentile > (microseconds) for "set" Workload: > 23:52:32.601 [main] INFO com.couchbase.roadrunner.RoadRunner - 50%:321 > 75%:365 95%:423 99%:5546 > 23:52:32.601 [main] INFO com.couchbase.roadrunner.RoadRunner - Elapsed: > 1499ms > 23:52:32.601 [main] INFO com.couchbase.roadrunner.RoadRunner - Shortest > Thread: 1487ms > 23:52:32.601 [main] INFO com.couchbase.roadrunner.RoadRunner - Longest > Thread: 1487ms > > On Thursday, January 30, 2014 8:35:11 PM UTC, Aliaksey Kandratsenka wrote: >> >> >> On Jan 30, 2014 12:32 PM, "asdf9898" <[email protected]> wrote: >> > >> > Hi: when I have two nodes: how do I disable replicas ? >> > >> > Replication is enabled and I cannot un-tick the option. >> > >> > And I cannot change the number of replicas from 1 to 0 >> >> Feel free to recreate bucket. >> >> But I believe your problem is more on client side. Your numbers look too >> small and thus there's not much reason to expect large speedup with more >> nodes. IMO >> > >> > Thanks >> > >> > >> > On Wednesday, January 29, 2014 8:22:57 PM UTC, asdf9898 wrote: >> >> >> >> Thanks Guys, >> >> >> >> I have a Windows machine with 4GB and a Windows machine with 8GB RAM. >> >> Both have 4 cpus. Regular hard drive. >> >> >> >> The total bucket in the cluster is 2x2GB. >> >> >> >> I'm loading 50,000 records which takes up maybe 100MB >> >> >> >> The insert time is between 5000 and 6000 records per second. >> >> A single node will give 6000 records but when two nodes I get just 5000 >> >> >> >> A single node can read back approx 22,000 records /s >> >> With two nodes I get maybe 14,000. This increased to 15,000 when I >> added the second URI of the second machine in the CouchbaseClient config. >> >> And increased to maybe 16,000 when I increased client threads to 32 >> >> >> >> I'm using the YCSB yahoo benchmark program. >> >> >> >> But I have found that using Elasticsearch 5 nodes on one machine and >> Hazelcast two nodes on one machine also causes a drop in throughput >> compared to a single node. >> >> >> >> By the way are you saying turning off replication could give a >> performance increase? That could be the problem? >> >> >> >> Regards, >> >> >> >> On Wednesday, January 29, 2014 7:46:33 PM UTC, Aliaksey Kandratsenka >> wrote: >> >>> >> >>> >> >>> >> >>> >> >>> On Wed, Jan 29, 2014 at 11:00 AM, asdf9898 <[email protected]> >> wrote: >> >>>> >> >>>> Hello >> >>>> >> >>>> I am new to nosql in general and would like some advice on couchbase >> and other similar technologies. >> >>>> >> >>>> I have done some performance testing of a number of technologies and >> find the same results with each. >> >>>> >> >>>> When I do x inserts on a single node I get better performance than >> when I add a second node. >> >>>> >> >>>> Read-performance is much better than insert performance but again a >> single node performs better than 2 nodes. >> >>>> >> >>>> I then thought that if I add the URIs in the buildConnection >> >>>> >> >>>> URI server = new URI(addresses); >> >>>> ArrayList<URI> serverList = new ArrayList<URI>(); >> >>>> serverList.add(server); >> >>>> CouchbaseClient client = new CouchbaseClient( >> >>>> serverList, "default", ""); >> >>>> >> >>>> In other words by explicitly telling the client about the second >> node it should cause double performance but sadly I just got maybe an extra >> 10% >> >>>> >> >>>> I then thought if I double the number of client threads because I >> now have two servers to target it would double performance but again no: >> just a slight increase. >> >>>> >> >>>> So in summary : would adding an extra node cause an increase or >> decrease in throughput (both reading and inserts) ? >> >>> >> >>> >> >>> Should increase (modulo tap load as Matt mentions). But in testing >> it, make sure that you don't jump to conclusions too quickly. I.e. when >> testing scalability I'd recommend looking at possible client-side >> contention too. I.e. if you have single client doing sequential ops, it >> obviously won't matter if you have 1 node or 100 nodes. All details matter. >> >>> >> > -- >> > You received this message because you are subscribed to the Google >> Groups "Couchbase" group. >> > To unsubscribe from this group and stop receiving emails from it, send >> an email to [email protected]. >> > For more options, visit https://groups.google.com/groups/opt_out. >> > -- You received this message because you are subscribed to the Google Groups "Couchbase" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/groups/opt_out.
