"4 times faster than protobuf" claim

2,698 views
Skip to first unread message

Marc Gravell

unread,
Jun 9, 2011, 7:07:18 AM6/9/11
to MessagePack Users
I'd like to point out that this claim is highly misleading:

- if profiling against the protobuf *format*, the only thing that
matters is the size
- otherwise, it only makes sense to make a claim in the context of a
very specific implementation

As an example, I'm the author of protobuf-net. I love serialization,
so I thought I'd see how the C# version compares.

My findings show that msgpack gives a marginally larger output (not
significantly larger), and is slower at both read (15% slower) and
write (50% slower).

That does *not* fit with a "4 times faster" claim *at all*.

My benchmark is shown here: http://pastie.org/2042052

I'm not trying to get into a p******g match - but simply: that claim
does not hold water (or is over-generalized to the point of
absurdity).

I'd love to hear back,

Marc Gravell
protobuf-net

James Gosnell

unread,
Nov 10, 2011, 3:34:37 PM11/10/11
to msg...@googlegroups.com
http://wiki.msgpack.org/display/MSGPACK/FAQ#FAQ-Whyhomepagestated4xfasterthanProtocolBuffer%3F

"The speed is compared with this benchmark. it measures the elapsed time of serializing and deserializing 200,000 target objects. The target object consists of the three integers and 512 bytes string.

This benchmark only uses C/C++ version of MessagePack implementation. The result of other language implementation will show the different result, because the maturity of language bindings currently varies a lot."

Andre Kaufmann

unread,
Nov 29, 2019, 8:24:24 AM11/29/19
to MessagePack Users
Hi,

just stumbled over this comment (ok a bit late) but hope still useful.
I had the same results, but then I remembered that the C# implementation of message pack uses IL intermediate compilation, meaning on the first access it will be slow.
Therefore I changed my benchmarks and didn't measure the first serialization, but the others. 
And then I got the result that Message Pack is faster. 

BR Andre
Reply all
Reply to author
Forward
0 new messages