thrift-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Diego Jancic <>
Subject Re: Performance issue
Date Thu, 16 Jul 2009 18:19:38 GMT
Thanks Michael. I was testing everything on similar machines.
I'll do a benchmark like yours and post the results. Maybe the results
on the windows machine with VS2008 will be similar, but I guess the
ones on the ubuntu machine with Mono will be 100 times slower.

Let me try again (maybe tomorrow). Do you want to send me the same
thrift and exe file you are using so I run them in Ubuntu?
The ubuntu machine has dual boot with Win XP, and is a Core 2 Duo, 4GB
RAM, 100Mb eth notebook. So I'll be able to compare running the same
in both linux and windows.


On Thu, Jul 16, 2009 at 2:12 PM, Michael Greene<> wrote:
> I don't know why this initial connection latency is there, and I'd be
> interested in helping you track it down if you come up with anything
> else.
> My usage pattern doesn't really run into this problem much.  If the
> connections are re-used at all it becomes much less of a problem.  For
> instance, with a slightly modified TestClient for better reporting I
> get the following results on a pretty mid-grade developer machine
> (Core 2 Duo, 4GB RAM, 100Mb eth):
>>ThriftTest.exe client -b -n 1
> Total time: 00:00:00.2219094 (hh:mm:ss)
> Time per run: 221ms
> Time per service call: 11.05ms
>>ThriftTest.exe client -b -n 10
> Total time: 00:00:00.4213149 (hh:mm:ss)
> Time per run: 42.1ms
> Time per service call: 2.105ms
>>ThriftTest.exe client -b -n 50
> Total time: 00:00:01.1194741 (hh:mm:ss)
> Time per run: 22.38ms
> Time per service call: 1.119ms
>>ThriftTest.exe client -b -n 500
> Total time: 00:00:08.9140619 (hh:mm:ss)
> Time per run: 17.828ms
> Time per service call: 0.8914ms
> This is on a Windows machine with VS2008, Mono results may vary -- we
> aren't using any secret Microsoft sauce though, and the results should
> be portable.
> Michael
> On Thu, Jul 16, 2009 at 11:22 AM, Diego Jancic<> wrote:
>> You're right, Thrift/Linux is not .net friendly ;-)
>> Anyway, I was not able to compile the C++ version of the example, but
>> I'll test with a sniffer on the weekend.
>> Thanks!
>> On Thu, Jul 16, 2009 at 12:59 PM, Jonathan Marcus<> wrote:
>>> I guess there aren't too many other C# users on this list, but I've also had
>>> a similar problem. I've posted about it before (on 4/24/2009), but there
>>> weren't any answers.
>>> I've used a sniffer and looked at the wire traffic for a simple test case,
>>> and I experience ~100ms latency from the first client->server data packet
>>> until the first response, on every single request.
>>> Has anybody has a successful C# implementation? If so, can you describe your
>>> setup, so that hopefully I can mimic it?
>>> --Jonathan
>>> Diego Jancic wrote:
>>>> Hi Guys!,
>>>> I'm exploring thrift for using it in c#, but I'm a little bit new to
>>>> Thrift,
>>>> Linux, and all that stuff.
>>>> Anyways, I compiled the example in trunk/test/csharp/ and ran it, but I
>>>> guess it's taking too long.
>>>> I've got 2 PCs, connected thru a 100 MB wired-network. One has Ubuntu and
>>>> the other has an WinXP
>>>> And the benchmarks are:
>>>> Server on Ubuntu <---> Client on Ubuntu  ............ 1.71 sec
>>>> Server on XP     <---> Client on XP      ............ 0.22 sec
Server on
>>>> Ubuntu <---> Client on XP      ............ 4.27 sec Server on XP
>>>> Client on Ubuntu  ............ 4.26 sec
>>>> I guess all are really slow times, even the XP-XP test. But I cannot
>>>> understand how can the Ubuntu-Ubuntu is working so slow.
>>>> The other configurations (Ubuntu-XP, and XP-Ubuntu) maybe are working so
>>>> slow due to the Ubuntu side, cause I don't guess a router with only 2
>>>> machines can make that work that way.
>>>> All tests are running over Mono, I'm gonna *try* to build another version
>>>> of
>>>> the example to test it.
>>>> Thanks for any tip,
>>>> Diego

View raw message