To Linux crunchers: lower credits issue
Message boards : Number crunching : To Linux crunchers: lower credits issue
| Author | Message | |
|---|---|---|
suguruhirahara
Forum moderator Volunteer tester 
Joined: Sep 13 06 Posts: 282 ID: 15 Credit: 56,614 RAC: 0 |
||
|
Hello,
|
||
ID:
1467 | Rating: 0
| rate:
/
|
||
|
|
||
|
ID:
1468 | Rating: 0
| rate:
/
|
||
|
Andre Kerstens
Forum moderator Project tester Volunteer tester 
Joined: Sep 11 06 Posts: 749 ID: 1 Credit: 15,199 RAC: 0 |
||
|
Thanks Suguru!
Hello, ____________ D@H the greatest project in the world... a while from now! |
||
|
ID:
1471 | Rating: 0
| rate:
/
|
||
|
suguruhirahara
Forum moderator Volunteer tester
Joined: Sep 13 06 Posts: 282 ID: 15 Credit: 56,614 RAC: 0 |
||
|
Hi all:)
|
||
|
ID:
1517 | Rating: 0
| rate:
/
|
||
|
@suguruhirahara, your calculation of 40% difference between Windows and Linux claims seems a bit high.
|
||
|
ID:
1677 | Rating: 0
| rate:
/
|
||
|
suguruhirahara
Forum moderator Volunteer tester
Joined: Sep 13 06 Posts: 282 ID: 15 Credit: 56,614 RAC: 0 |
||
|
The new version (5.6.4 to 5.7.5) of BOINC for linux to be tested became available on
the download page
. It implemented new feature of 'advanced GUI only', but as far as I checked the benchmark system was improved a little though it's not perfect. Try it if you want to.
|
||
|
ID:
1760 | Rating: 0
| rate:
/
|
||
|
@suguruhirahara, your calculation of 40% difference between Windows and Linux claims seems a bit high. As you'd know, CPU time is one factor in the way BOINC claims credit. We also know the app is much faster under Linux. Example 1: My Athlon XP 3000+ Windows: approx 23,500 seconds for 83 credits (12.7 per hour) Linux: approx. 14,000 seconds for 28 credits (7.2 per hour) Example 2: My Pentium 4/3.4 with HT Windows: approx 35,500 seconds for 86 credits (8.7 per hour per thread) Linux: approx. 20,000 seconds for 29 credits (5.2 per hour per thread) The total claimed per work unit under Linux is about 33% of the claim under Windows. But it takes about 58% of CPU time. On this project at least, per hour of CPU time, the Linux host claims about 55-60% of the Windows host, not 25-30%. Still a big difference (and a BOINC problem), but not as bad as when you look at it purely from a credit per work unit perspective. ____________ 
Join the #1 Aussie Alliance on Docking@Home |
||
|
ID:
1768 | Rating: 0
| rate:
/
|
||
|
The new version (5.6.4 to 5.7.5) of BOINC for linux to be tested became available on the download page . It implemented new feature of 'advanced GUI only', but as far as I checked the benchmark system was improved a little though it's not perfect. Try it if you want to. Benchmark indeed looks higher... 5.7.5: Measured floating point speed 1301.74 million ops/sec Measured integer speed 2277.85 million ops/sec 5.4.9: Measured floating point speed 993.78 million ops/sec Measured integer speed 1740.58 million ops/sec Result... my linux host is now claiming approx 29-30 credits/13.400 seconds. (5.4.9 approx 22-23/13.400) As far as I can tell this is close to bench-vs-claim of my Windows host. Besides from that the 5.7.5 GUI looks a lot better... ;-) ____________ 
|
||
|
ID:
1775 | Rating: 0
| rate:
/
|
||
|
suguruhirahara
Forum moderator Volunteer tester
Joined: Sep 13 06 Posts: 282 ID: 15 Credit: 56,614 RAC: 0 |
||
|
Besides from that the 5.7.5 GUI looks a lot better... ;-) Yes, this is one of the key update features. Boinc manager 5.4.9 is frankly not...good one. The new one is quite better:) ____________ 
I'm a volunteer participant; my views are not necessarily those of Docking@Home or its participating institutions. |
||
|
ID:
1776 | Rating: 0
| rate:
/
|
||
|
5.7.5 linux is closer to windoze for benchmark but hard to believe
A
&
B
are the same machine. with 800 less mflops &300 less mips linux still appears to be over 20% faster
|
||
|
ID:
1880 | Rating: 0
| rate:
/
|
||
|
David Ball
Forum moderator Volunteer tester 
Joined: Sep 18 06 Posts: 274 ID: 115 Credit: 1,634,401 RAC: 0 |
||
|
5.7.5 linux is closer to windoze for benchmark but hard to believe A & B are the same machine. with 800 less mflops &300 less mips linux still appears to be over 20% faster I think that has something to do with the D@H project client writing to disk so much. Linux seems to handle that better by buffering and delaying the actual writes longer. IIRC, someone actually took a Windows XP machine, loaded VMware on it, and found that a WU ran faster in Linux on VMware on WinXP, than it ran natively on just WinXP on the same machine. -- David |
||
|
ID:
1887 | Rating: 1
| rate:
/
|
||
|
IIRC, someone actually took a Windows XP machine, loaded VMware on it, and found that a WU ran faster in Linux on VMware on WinXP, than it ran natively on just WinXP on the same machine. My 3.4GHz Pentium 4 with HT took 5.5 hours for a single WU in Linux under VMWare (while also running BOINC in the host O/S). Running Docking in the host O/S (Windows 2003 SBS), work units took on average about 10 hours. Whether this is related to disk performance, I don't know - it was using the same hard-drive and (presumably) still had to patch it through the host O/S. |
||
|
ID:
1893 | Rating: 0
| rate:
/
|
||
|
David Ball
Forum moderator Volunteer tester
Joined: Sep 18 06 Posts: 274 ID: 115 Credit: 1,634,401 RAC: 0 |
||
|
IIRC, someone actually took a Windows XP machine, loaded VMware on it, and found that a WU ran faster in Linux on VMware on WinXP, than it ran natively on just WinXP on the same machine. I think that Linux just stuffs the data the application writes into a buffer and returns success to the program that did the writing. Then Linux has a kernel thread which does the actual writes to disk about every 5 seconds. This lets the actual disk writes happen in parallel with the application rather than the application waiting for each of them to complete. I'm not sure of the exact windows algorithm but it seems like windows not only does the write while making the application wait for the write to complete, but charges that time to the application. Even if VMware goes through Windows for the writes by the VM Linux kernel thread, it must be done by a different VMware Windows thread and occur in parallel with the Linux application running. -- David |
||
|
ID:
1895 | Rating: 0
| rate:
/
|
||
|
IIRC, someone actually took a Windows XP machine, loaded VMware on it, and found that a WU ran faster in Linux on VMware on WinXP, than it ran natively on just WinXP on the same machine. Is the vmware actual wallclock time or reported cpu time? I'm pretty sure that this was looked at on another project and it was found that the cputime reported under vmware was substantially different even when "nothing else" was running on vmware or the host system. |
||
|
ID:
1902 | Rating: 0
| rate:
/
|
||
|
Is the vmware actual wallclock time or reported cpu time? I'm talking about reported CPU time in both cases. No idea how long it took in real time. |
||
|
ID:
1903 | Rating: 0
| rate:
/
|
||
|
All this talk about newer BOINC versions fixing the disparity between Linux and Windows got me curious. I downloaded BOINC 5.8.1 (development version) to my 3.4GHz Pentium 4 (HT) which is dual boot now and I ran benchmarks a few times, with mixed results.
|
||
|
ID:
1973 | Rating: 0
| rate:
/
|
||
|
I also upgraded my Dual Athlon XP 2 GHz running Ubuntu 6.10 to the new BOINC version 5.8.1 (development version) now and the claimed credits have increased significantly:
|
||
|
ID:
2025 | Rating: 0
| rate:
/
|
||
|
suguruhirahara
Forum moderator Volunteer tester
Joined: Sep 13 06 Posts: 282 ID: 15 Credit: 56,614 RAC: 0 |
||
|
Hello all,
|
||
|
ID:
2330 | Rating: 0
| rate:
/
|
||
Message boards : Number crunching : To Linux crunchers: lower credits issue
Database Error: The MySQL server is running with the --read-only option so it cannot execute this statement
array(3) {
[0]=>
array(7) {
["file"]=>
string(47) "/boinc/projects/docking/html_v2/inc/db_conn.inc"
["line"]=>
int(97)
["function"]=>
string(8) "do_query"
["class"]=>
string(6) "DbConn"
["object"]=>
object(DbConn)#23 (2) {
["db_conn"]=>
resource(108) of type (mysql link persistent)
["db_name"]=>
string(7) "docking"
}
["type"]=>
string(2) "->"
["args"]=>
array(1) {
[0]=>
&string(51) "update DBNAME.thread set views=views+1 where id=103"
}
}
[1]=>
array(7) {
["file"]=>
string(48) "/boinc/projects/docking/html_v2/inc/forum_db.inc"
["line"]=>
int(60)
["function"]=>
string(6) "update"
["class"]=>
string(6) "DbConn"
["object"]=>
object(DbConn)#23 (2) {
["db_conn"]=>
resource(108) of type (mysql link persistent)
["db_name"]=>
string(7) "docking"
}
["type"]=>
string(2) "->"
["args"]=>
array(3) {
[0]=>
object(BoincThread)#3 (16) {
["id"]=>
string(3) "103"
["forum"]=>
string(1) "2"
["owner"]=>
string(2) "15"
["status"]=>
string(1) "0"
["title"]=>
string(39) "To Linux crunchers: lower credits issue"
["timestamp"]=>
string(10) "1169885850"
["views"]=>
string(4) "1911"
["replies"]=>
string(2) "17"
["activity"]=>
string(20) "1.4308522806097e-125"
["sufferers"]=>
string(1) "0"
["score"]=>
string(1) "0"
["votes"]=>
string(1) "0"
["create_time"]=>
string(10) "1163755172"
["hidden"]=>
string(1) "0"
["sticky"]=>
string(1) "0"
["locked"]=>
string(1) "0"
}
[1]=>
&string(6) "thread"
[2]=>
&string(13) "views=views+1"
}
}
[2]=>
array(7) {
["file"]=>
string(63) "/boinc/projects/docking/html_v2/user/community/forum/thread.php"
["line"]=>
int(184)
["function"]=>
string(6) "update"
["class"]=>
string(11) "BoincThread"
["object"]=>
object(BoincThread)#3 (16) {
["id"]=>
string(3) "103"
["forum"]=>
string(1) "2"
["owner"]=>
string(2) "15"
["status"]=>
string(1) "0"
["title"]=>
string(39) "To Linux crunchers: lower credits issue"
["timestamp"]=>
string(10) "1169885850"
["views"]=>
string(4) "1911"
["replies"]=>
string(2) "17"
["activity"]=>
string(20) "1.4308522806097e-125"
["sufferers"]=>
string(1) "0"
["score"]=>
string(1) "0"
["votes"]=>
string(1) "0"
["create_time"]=>
string(10) "1163755172"
["hidden"]=>
string(1) "0"
["sticky"]=>
string(1) "0"
["locked"]=>
string(1) "0"
}
["type"]=>
string(2) "->"
["args"]=>
array(1) {
[0]=>
&string(13) "views=views+1"
}
}
}
query: update docking.thread set views=views+1 where id=103