Have you tested the bandwidth between the two machines using Netperf ? You handling of streaming using isReady/onReady is good, we have an improvement planned to add 'corking' https://github.com/grpc/grpc-java/issues/994 which can substantially improve throughput when dealing with large numbers of small messages in a stream.
One thing I notice between your two APIs is that in the more structure one you send the column name for every column value which seems quite inefficient On Tue, Oct 4, 2016 at 1:31 AM, Avinash Dongre <[email protected]> wrote: > Hi Louis, > > Thanks, I have made the changes and I see good numbers with async client. > I have committed the change in my repo. > > But I still see very low data transfer rate. What I have observed is that > when I have big data say ( 1024 bytes ) Then I get rate on the same machine > around 900-1000 MegaBytes/seconds. > > Thanks > Avinash > > Num Of Rows -> 2000000 > Total Data Size -> 512000000 > > Time -> 7.484403756 Seconds > Data Rate -> 65.20225470316008 MBps > > Time -> 5.776407626 Seconds > Data Rate -> 84.48157256137519 MBps > > Time -> 6.010102214 Seconds > Data Rate -> 81.19662239075524 MBps > > Time -> 5.401799312 Seconds > Data Rate -> 90.34026845757057 MBps > > Time -> 5.352594307 Seconds > Data Rate -> 91.17074300994655 MBps > > > > > On Tuesday, October 4, 2016 at 4:48:50 AM UTC+5:30, Louis Ryan wrote: >> >> I see you're using a blocking stub to make the streaming calls. I suggest >> you switch to using an async callback instead here >> >> https://github.com/davinash/grpc-bench/blob/master/bench/src >> /main/java/io/adongre/grpc/formatted/ScanFormattedClient.java#L35 >> >> E.g. >> >> asyncStub.formatedScan(ScanRequest.newBuilder()..., new StreamObserver() >> { ... public void onValue( ScanFormattedResponse response) { .... do >> stuff with response. i.e. replace the while loop in your code } }); >> .setNumOfColumns(NUMBER_OF_COLUMNS) >> .setNumOfRows(NUM_OF_ROWS) >> .setSizeOfEachColumn(SIZE_OF_EACH_COLUMN) >> .build()); >> >> On Sun, Oct 2, 2016 at 3:16 AM, Avinash Dongre <[email protected]> >> wrote: >> >>> Thanks Louis, For helping me here >>> >>> Sorry for delayed response >>> >>> I am sure I am doing something wrong in my code, but could not figure >>> out yet. >>> >>> I have checked-in my benchmark project here at >>> https://github.com/davinash/grpc-bench >>> >>> Raw Server , client and impl can be found at https://github.com/davinash >>> /grpc-bench/tree/master/bench/src/main/java/io/adongre/grpc/raw >>> >>> Formatted Server, client and Impl can be found at >>> https://github.com/davinash/grpc-bench/tree/master/bench/ >>> src/main/java/io/adongre/grpc/formatted >>> >>> With this above code for Raw I get following numbers >>> >>> Num Of Rows -> 2000000 >>> Time -> 68.468212235 Seconds >>> Total Data Size -> 65536000000 >>> Data Rate -> 912.8323635132227 MBps >>> >>> >>> >>> And with formatted one I get following numbers. >>> >>> Num Of Rows -> 2000000 >>> Time -> 358.549532039 Seconds >>> Total Data Size -> 81952000000 >>> Data Rate -> 217.97546228982097 MBps >>> >>> >>> Thanks >>> Avinash >>> >>> >>> >>> >>> >>> On Monday, September 26, 2016 at 10:08:47 PM UTC+5:30, Louis Ryan wrote: >>>> >>>> I see that you're using a steaming RPC. Can you send the code from the >>>> server and client. How you interact with flow control can greatly affect >>>> performance >>>> >>>> -louis (from phone) >>>> >>>> On Sep 26, 2016 3:57 AM, "Avinash Dongre" <[email protected]> wrote: >>>> >>>> Hi All, >>>> Please help. >>>> >>>> Thanks >>>> Avinash >>>> >>>> >>>> On Saturday, September 24, 2016 at 12:01:02 PM UTC+5:30, Avinash Dongre >>>> wrote: >>>>> >>>>> >>> Now I get around 130-135 MegaBytes/Seconds Speed. >>>>> >>>>> This result is on the Same Machine. i.e. gRPC Client and gRPC Servers are >>>>> running on the same machine. >>>>> >>>>> >>>>> >>>>> On Saturday, September 24, 2016 at 11:59:53 AM UTC+5:30, Avinash >>>>> Dongre wrote: >>>>>> >>>>>> Hi All, >>>>>> >>>>>> Thanks all for replying/resolving my previous doubts. I am planning >>>>>> to use gRPC for our project. I have following question related to >>>>>> performance. >>>>>> >>>>>> If I have following proto definition. >>>>>> >>>>>> option optimize_for = SPEED; >>>>>> message ScanRow { >>>>>> repeated bytes row = 1; >>>>>> } >>>>>> message ScanResult { >>>>>> repeated ScanRow row = 1; >>>>>> } >>>>>> message ScanRequest { >>>>>> int32 numOfColumns = 1; >>>>>> int32 sizeOfEachColumn = 2; >>>>>> int64 numOfRows = 3; >>>>>> int32 batchSize = 4; >>>>>> } >>>>>> service ScanService { >>>>>> rpc Scan (ScanRequest) returns (stream ScanResult) {} >>>>>> } >>>>>> >>>>>> I get around 900-1000 MegaBytes/Seconds on Single Machine and across >>>>>> two different Physical Machines I am getting 100-110 MegaBytes/Seconds >>>>>> speed. >>>>>> >>>>>> If I change above proto definition to following. >>>>>> >>>>>> option optimize_for = SPEED; >>>>>> message ColumnValues { >>>>>> bytes columnName = 1; >>>>>> bytes columnValue = 2; >>>>>> } >>>>>> message ScanRow { >>>>>> int64 rowId = 1; >>>>>> int64 timeStamp = 2; >>>>>> repeated ColumnValues columnValue = 3; >>>>>> } >>>>>> message ScanResult { >>>>>> repeated ScanRow row = 1; >>>>>> } >>>>>> message ScanRequest { >>>>>> int32 numOfColumns = 1; >>>>>> int32 sizeOfEachColumn = 2; >>>>>> int64 numOfRows = 3; >>>>>> int32 batchSize = 4; >>>>>> } >>>>>> service ScanService { >>>>>> rpc Scan (ScanRequest) returns (stream ScanResult) {} >>>>>> } >>>>>> >>>>>> Now I get around 130-135 MegaBytes/Seconds Speed. >>>>>> >>>>>> >>>>>> Why it is slow with kind of multi-level proto files. Is there any >>>>>> Serialization/De-Serialization overhead with this ? >>>>>> >>>>>> >>>>>> Thanks >>>>>> >>>>>> Avinash >>>>>> >>>>>> >>>>>> -- >>>> You received this message because you are subscribed to the Google >>>> Groups "grpc.io" group. >>>> To unsubscribe from this group and stop receiving emails from it, send >>>> an email to [email protected]. >>>> To post to this group, send email to [email protected]. >>>> Visit this group at https://groups.google.com/group/grpc-io. >>>> To view this discussion on the web visit https://groups.google.com/d/ms >>>> gid/grpc-io/cc74ecdf-3eac-4475-bad0-23fd66ee03b2%40googlegroups.com >>>> <https://groups.google.com/d/msgid/grpc-io/cc74ecdf-3eac-4475-bad0-23fd66ee03b2%40googlegroups.com?utm_medium=email&utm_source=footer> >>>> . >>>> >>>> For more options, visit https://groups.google.com/d/optout. >>>> >>>> >>>> -- >>> You received this message because you are subscribed to the Google >>> Groups "grpc.io" group. >>> To unsubscribe from this group and stop receiving emails from it, send >>> an email to [email protected]. >>> To post to this group, send email to [email protected]. >>> Visit this group at https://groups.google.com/group/grpc-io. >>> To view this discussion on the web visit https://groups.google.com/d/ms >>> gid/grpc-io/76dcf498-35e2-4e5a-ab50-92e708cdb808%40googlegroups.com >>> <https://groups.google.com/d/msgid/grpc-io/76dcf498-35e2-4e5a-ab50-92e708cdb808%40googlegroups.com?utm_medium=email&utm_source=footer> >>> . >>> >>> For more options, visit https://groups.google.com/d/optout. >>> >> >> -- > You received this message because you are subscribed to the Google Groups " > grpc.io" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at https://groups.google.com/group/grpc-io. > To view this discussion on the web visit https://groups.google.com/d/ > msgid/grpc-io/99752304-bbb3-4a6b-8e43-225629408e76%40googlegroups.com > <https://groups.google.com/d/msgid/grpc-io/99752304-bbb3-4a6b-8e43-225629408e76%40googlegroups.com?utm_medium=email&utm_source=footer> > . > > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "grpc.io" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at https://groups.google.com/group/grpc-io. To view this discussion on the web visit https://groups.google.com/d/msgid/grpc-io/CADQ0XY0cz1WTSjFNDQGDo_C0%2BooYb1xNtJZnDAiu%3D3MQkbmtwA%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
