Home Comments

bench tests

alfredtalfredt Member Alfred TasconOrganization: NoneProject: None
I've just ran the Freeze/bench tests, and wondered why the reads are taking so long?

I am getting inserts ~0.5 ms each, while reads are ~4.8ms each.

And it doesn't matter if it is reading 1 record or 1,000 records.

what's the deal?




  • bernardbernard Jupiter, FLAdministrators, ZeroC Staff Bernard NormierOrganization: ZeroC, Inc.Project: Ice ZeroC Staff
    Hi Alfred,

    Good question! I ran the Freeze/bench C++ demo using an optimized Linux build, and got similar results: for the IntIntMap, inserts take 0.06 to 0.14ms, while reads take 0.35ms (yes, I have a fast box :D).

    Most of the read time is consumed by the XML demarshalling; I switched to binary encoding (by adding --binary to slice2freeze) and got a more reasonable 0.03ms per read.
    Unfortunately switching to binary encoding on Linux triggered a strange compilation problem in Struct1Class1MapValueCodec::write(). I am investigating it now.

    Thanks for raising this issue! We will address this XML demarshalling performance issue in the next Ice release.

  • alfredtalfredt Member Alfred TasconOrganization: NoneProject: None
    I also switched slice2freeze to --binary under VC7.1 and got the same error.

    Managed to compile it and also got desent read times of about a third the write times in the IntIntMap test.

    It only ran the first two tests then raised a nullhandle exception at the Struct1Class1Map read test.

Sign In or Register to comment.