My software RAID-0 reads slower than it writes?

[funkahdafi]

Hi,

I am using two disks in a RAID-0 array, configured via Disk Util. Works pretty good. Now I did some basic "benchmarking" just to find out it writes way faster than it reads. Shouldn't it be the other way around? Here are some results:

reptilehouse:~ sascha$ time dd if=/dev/zero of=/Volumes/HD1/testfile bs=1024k count=16384
16384+0 records in
16384+0 records out
17179869184 bytes transferred in 87.676360 secs (195946424 bytes/sec)

real    1m27.774s
user    0m0.065s
sys     0m22.029s


reptilehouse:~ sascha$ time dd of=/dev/null if=/Volumes/HD1/testfile bs=1024
16777216+0 records in
16777216+0 records out
17179869184 bytes transferred in 99.680770 secs (172348881 bytes/sec)

real    1m39.696s
user    0m22.333s
sys     1m9.159s

So that's roughly 200 MB/sec. when writing vs. around 170 MB/sec. when reading. That's kind of odd. This is reproducable by the way and happens all the time.

Any thoughts on this?

Thanks

 

[funkahdafi]

Anyone?

 

[Guiyon]

Have you double checked your commands? The write test is using 1MB blocks while the read test looks like it is using 1KB blocks.

 

[funkahdafi]

Have you double checked your commands? The write test is using 1MB blocks while the read test looks like it is using 1KB blocks.

Good point. You are right. I just tried again by using bs=1024k when reading, but I am still getting the same results. Reading is still a lot slower. Last time I tried write=195MB/sec. vs. read=157MB/sec.

 

[ElectricSheep]

Its likely that what you are seeing here are the effects of write-back caching. Blocks you think are being written to disk may be cached in RAM before a flush operation. This can artificially inflate write benchmarks with tools like dd.

 

[funkahdafi]

Its likely that what you are seeing here are the effects of write-back caching. Blocks you think are being written to disk may be cached in RAM before a flush operation. This can artificially inflate write benchmarks with tools like dd.

Makes sense.

However, I chose a filesize twice as large as my physical memory. And my hard drives have 1 throughput of 100 MB/sec (single). So I would expect to see 200 MB/sec. in raid 0 when reading.