Hi Evan,
Performance is something that will depend on the nature of your workload
quite a lot, so it's hard to give advice without knowing more detail. Here
are some recommendations that we tend to give:
- Using arenas can reduce the cost of memory allocation. This can be
especially impactful during parsing (deserializing).
- https://protobuf.dev/reference/cpp/arenas/
- Protobuf reflection is slow -- try to avoid it in performance
sensitive code.
- string fields are UTF-8 validated. If you don't need UTF-8 validation,
prefer to use bytes instead.
- Deeply nested messages can result in a lot of pointer indirection. In
some cases, you might be able to come up with a wire-compatible message
format that is shallower. However, this may come at the cost of flexibility.
- Serialization performance is O(number of fields). If you have a large
message with 1000s or 10000s of fields, serializing it can be noticeably
slower than serializing a message with a handful of fields.
We also have best practices published here:
https://protobuf.dev/programming-guides/dos-donts/. It's not focused on
performance though.
And please keep in mind that we are continually evolving the protobuf
implementation, so anything we're saying here isn't a substitute for
running your own benchmarks. :)
Cheers,
Tony
On Monday, September 9, 2024 at 5:23:37 PM UTC-4 [email protected] wrote:
> Hi Tony,
>
> Thank you for the clarification!
>
> Do you have any recommendations or best practices for improving
> serialization/deserialization performance specifically for Protobuf in C++?
> I'd appreciate any insights or optimizations you or anyone can suggest.
>
> Best regards,
> Evan
>
> On Monday, September 9, 2024 at 1:52:01 PM UTC-7 Tony Liao wrote:
>
>> Hi Evan,
>>
>> The protobuf C++ implementation is separate from UPB and has its own set
>> of optimizations driven by Google's own workloads.
>>
>> I can say that for Google's C++ workloads, we have not observed better
>> performance with UPB. For this reason, we are still shipping with the
>> native implementation for C++.
>>
>> -Tony
>>
>> On Friday, September 6, 2024 at 7:08:09 PM UTC-4 [email protected] wrote:
>>
>>> Hi everyone,
>>>
>>> We're currently using Protobuf 3.14 with C++ and Python in our project
>>> and are looking to improve serialization/deserialization performance. I
>>> recently tried Protobuf 3.24 and noticed performance improvements in
>>> Python, likely due to the use of UPB.
>>>
>>> I have a couple of questions:
>>>
>>> 1. Does UPB also provide performance improvements for C++?
>>> 2. If so, in which version of Protobuf was UPB introduced for C++?
>>>
>>> Thanks in advance for your help!
>>>
>>> Best regards,
>>> Evan
>>>
>>
--
You received this message because you are subscribed to the Google Groups
"Protocol Buffers" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/protobuf/e15db3f5-7761-422c-8b32-5ceea719a26cn%40googlegroups.com.
