0

m using pelops to talk to cassandra and m wondering with its slower performance on server than local when m trying to insert the 1 GB of data .I m trying a bulk insert that is m creating a batch in mutator.writecolumn of around 2000 columns and using mutator.execute m inserting .1 GB of data processing taking more than 25 mins. i want to know if there any solution for this or this is the normal performance of cassandra?

have created the cluster of 3 nodes. cassandra 7.4 pelops thrift 0.5

thanks and regards samarth

jbellis
  • 19,347
  • 2
  • 38
  • 47
samarth
  • 3,866
  • 7
  • 45
  • 60

2 Answers2

1

You don't give many details, but if I were to guess I would say you are probably doing single-threaded inserts which means you're limited to < 1 Cassandra core.

jbellis
  • 19,347
  • 2
  • 38
  • 47
  • no i am doing the multithreaded insert,but inbetween i face out of sequence exception or unable to write to null output streams any idea why m facing this? – samarth Aug 04 '11 at 14:16
  • Sounds like you're using the same Pelops connection from multiple threads. You need to use separate connection objects. – jbellis Aug 05 '11 at 15:38
0

What is your write parameter for ... hm, I forgot teh terminology casandra sues. There is a parameter you can add to a write command, that specifies on how many nodes the write must have happend to be considered "successfull". Obviously if you run localy that parameter liekly is 1 (I think you can set that in the config file of the DB), if you have a cluster of 3 nodes you likely have set it to 3 or to 2. Ofc it takes much longer to write adn verify that it is written if that is the case.

Angel O'Sphere
  • 2,642
  • 20
  • 18
  • Not so, because Cassandra always writes all the replicas when your cluster is healthy. ConsistencyLevel is more about "how many replicas need to be healthy for me to proceed." – jbellis Jul 26 '11 at 14:59
  • Yes, ofc!! But that determiens how long it takes that your "write call" returns! So when you have a local DB with only 1 node yourn write call might take 5 seconds, but the same call to a cluster fo 3 nodes will take 15 secodns if you demand consistency level 3!! – Angel O'Sphere Jul 27 '11 at 14:08
  • True, that it is done parallel, I forgot about that. Nevertheless it takes longer than writing once ... did you find a possible solution to your problem? Perhaps it is only network related? Or the nodes use inapprobriated JVM starting parameters (not enough memory)? – Angel O'Sphere Jul 29 '11 at 07:44
  • is there any issue if i use multithreaded invironment for writing to cassandra.will it break the socket?m getting out of sequence and unable to write to cassandra exception\ – samarth Aug 04 '11 at 14:19