[
Lists Home |
Date Index |
Thread Index
]
Dave Winer says this on the dist-app list:
>To form an opinion on this you'd have to have an application in mind and
>then do some performance comparisons under realistic loads. My intuition
>says it doesn't matter much how fat XML is for protocols -- the Internet and
>LANs have scaled to push around lots of images and MP3s, which are huge
>compared to SOAP or XML-RPC messages.
This is actually wrong, though intuitively you'd believe it to be so. The way
TCP/IP generally works, large files tend to get transferred fairly
efficiently, as do small RPC calls.... and smallish things tend to get
penalized.
In order to make a fair comparison, you have to compare the cost of making an
RPC call, say a function that takes 2 strings/integers and computes a result.
There is a penalty here for even increasing the overall message size by one
or two packets.
When I did my tests a few years ago, I found RMI to be slower than hand-coded
XML-RPC calls... apparently because JAVA serialization was slow. However,
XML-RPC was, in general *much* slower than hand-coded RPC.
Despite speed there are the obvious network management issues of resource
planning etc.
Anyway, this isn't a flame against SOAP per se., so much as a note that it
isn't anywhere nearly as simple as people might think.
|
