Once again, because this is for every mail account... Please look at the messages below. If anyone needs the full snapshot of yourkit, contact me.
---------- Forwarded message ---------- From: Thomas Jungblut <[email protected]> Date: 2011/4/22 Subject: Fwd: [Share] Trunk is little slower than before. Since the apache mail daemon only allows 1000000 bytes I'll send it to you directly. Look at the message below. Best regards, Thomas ---------- Forwarded message ---------- From: Thomas Jungblut <[email protected]> Date: 2011/4/22 Subject: Re: [Share] Trunk is little slower than before. To: [email protected] Results: I used "bench 524288 100 200". The screenshot of method invocation profiling is in the attachment. Funny, that the execution time of bsp() is just 50% of the time. Most of this time is spent waiting. Take a look at the invocation time of PingInputStream.read(byte[]). It takes 39% of the whole execution time (!). Hadoop's source is saying this for this class: /** This class sends a ping to the remote side when timeout on > * reading. If no failure is detected, it retries until at least > * a byte is read. */ > If you own yourkit, the snapshot is in the attachment too. Used version 9.5.5. Cu 2011/4/22 Thomas Jungblut <[email protected]>: > Oh sorry all, this was a private massage of Miklos, but I didn't see > this because of google mails conversation threading. So Miklos that > was your message ;D > > 2011/4/22 Thomas Jungblut <[email protected]>: >> Hi all, >> >> my favorite profiler is YourKit. http://yourkit.com/ (This is not an >> advertisement :D) But for multithreading purposes it is better than >> the others. You have a good overview of threads that are blocked by >> others and so on. >> But it is really expensive, there is an option if you are developing >> for OpenSource that you'll get a free licence. For testing you have I >> believe 14 days free testing phase (with all options and no >> restrictions). You could try out. >> >> +1 for recording times ;) >> >> >> 2011/4/22 Chia-Hung Lin <[email protected]>: >>> In the wiki page, there are several information recorded, such as time >>> elapsed, message size, superstep counts, and number of message sent >>> (per superstep). In addition to this, is there any chance other >>> information can be recorded as well? For instance, how long does it >>> take to achieve barrier synchronization? Or time elapsed when >>> dispatching tasks. >>> >>> 2011/4/22 Edward J. Yoon <[email protected]>: >>>> +1 >>>> >>>> Sent from my iPad >>>> >>>> On Apr 22, 2011, at 7:13 PM, Thomas Jungblut < [email protected]> wrote: >>>> >>>>> This is MUCH slower. I'll profile it later. >>>>> And IMHO we should remove the random component of the benchmark. >>>>> >>>>> 2011/4/22 Edward J. Yoon <[email protected]>: >>>>>> Hi all, >>>>>> >>>>>> I ran the RandBench of current trunk on the same condition (16 nodes, >>>>>> 10G network): >>>>>> >>>>>> http://wiki.apache.org/hama/Benchmarks >>>>>> >>>>>> 0.2.0 version -> current TRUNK (arguments) >>>>>> -- >>>>>> 7.461 seconds -> 12.445 seconds (500, 100, 16) >>>>>> 84.489 seconds -> 147.521 seconds (500, 100, 512) >>>>>> >>>>>> 13.679 seconds -> 18.419 seconds (10000, 5, 16) >>>>>> 339.608 seconds -> 402.632 seconds (10000, 5, 512) >>>>>> >>>>>> Trunk is little slower than before! >>>>>> >>>>>> -- >>>>>> Best Regards, Edward J. Yoon >>>>>> http://blog.udanax.org >>>>>> http://twitter.com/eddieyoon >>>>>> >>>>> >>>>> >>>>> >>>>> -- >>>>> Thomas Jungblut >>>>> Berlin >>>>> >>>>> mobile: 0170-3081070 >>>>> >>>>> business: [email protected] >>>>> private: [email protected] >>>> >>> >> >> >> >> -- >> Thomas Jungblut >> Berlin >> >> mobile: 0170-3081070 >> >> business: [email protected] >> private: [email protected] >> > > > > -- > Thomas Jungblut > Berlin > > mobile: 0170-3081070 > > business: [email protected] > private: [email protected] > -- Thomas Jungblut Berlin mobile: 0170-3081070 business: [email protected] private: [email protected] -- Thomas Jungblut Berlin mobile: 0170-3081070 business: [email protected] private: [email protected] -- Thomas Jungblut Berlin mobile: 0170-3081070 business: [email protected] private: [email protected]
