[DRBD-user] Re: Why is drbd 0.7.22 slow on Opteron 240 EE?

Leroy van Logchem leroy.vanlogchem at wldelft.nl
Wed Nov 22 11:48:49 CET 2006

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


Maurice Volaski wrote:
>> and then to benchmark:
>>  ./dm -x -a 0 -s 4g -b 20m -m -y -p -o /maurice/bonnie/hello
> Here are the results for the two computers
> primary (faster one)
> 29.34 MB/sec (4294967296 B / 02:19.620474)
> secondary (slower one)
> 34.57 MB/sec (4294967296 B / 01:58.497313)
> And why do they contradict the much faster results from dd?

When using dd it's only writen to page-cache it doesnt it the disk if 
you got enough
memory. My 2-cents on this:

$ time dd if=/dev/zero of=hello bs=400M oflag=sync count=1 && time sync
1+0 records in
1+0 records out
419430400 bytes (419 MB) copied, 8.66923 seconds, 48.4 MB/s
real    0m8.808s
user    0m0.000s
sys     0m1.548s
real    0m0.337s  <-- flush cache to disk using SYNC(1)
user    0m0.000s
sys     0m0.004s +
------------------------
Total 9.1 sec

$ time dd if=/dev/zero of=hello bs=400M count=1 && time sync
1+0 records in
1+0 records out
419430400 bytes (419 MB) copied, 3.17573 seconds, 132 MB/s <- This is 
fictional imo
real    0m3.307s
user    0m0.000s
sys     0m1.324s
real    0m5.650s   <-- sync
user    0m0.000s
sys     0m0.204s +
------------------------
Total 8.9 sec
 
The write barrier of Protocol 'C' wants it really on disk -bypassing 
pagecache-, can someone confirm this?



More information about the drbd-user mailing list