• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Read less than write

P

planum

n00b
Joined
Mar 10, 2011
Messages
17
Hi, I am new to the forum and have a zfs specific question
I have a 12 bay array with 1tb drives in a raid-z2 configuration newly setup.
This is a result of iostat with a ~1100 file copy operation from on directory to another on the same machine (so no network slow downs).

rlksun@proxima:~# zpool iostat -v
capacity operations bandwidth
pool alloc free read write read write
------------------------- ----- ----- ----- ----- ----- -----
rpool 12.8G 61.7G 0 12 60.4K 127K
c2t0d0s0 12.8G 61.7G 0 12 60.4K 127K
------------------------- ----- ----- ----- ----- ----- -----
tank 3.25T 7.62T 127 190 14.8M 20.1M
raidz2 3.25T 7.62T 127 190 14.8M 20.1M
c1t50014EE05667227Ed0 - - 25 31 1.60M 2.03M
c1t5000C50030EF862Ed0 - - 22 30 1.63M 2.03M
c1t5000C50030EF86D2d0 - - 23 30 1.58M 2.02M
c1t5000C50030EF8ABDd0 - - 23 30 1.61M 2.03M
c1t5000C50030EEC711d0 - - 23 30 1.63M 2.02M
c1t5000C50030EDEFE6d0 - - 22 29 1.58M 2.02M
c1t5000C50030ED381Bd0 - - 22 30 1.61M 2.03M
c1t5000C50030E7313Bd0 - - 23 30 1.63M 2.03M
c1t5000C50030E303A4d0 - - 22 30 1.58M 2.02M
c1t5000C50030E4E235d0 - - 22 30 1.61M 2.03M
c1t5000C50030E4E5C2d0 - - 23 30 1.63M 2.02M
c1t5000C50022AF0A29d0 - - 22 29 1.59M 2.02M
------------------------- ----- ----- ----- ----- ----- -----

My question is simple, why would the write speeds be faster than the read speeds? Unless I am misinterpreting the output.
Thanks in advance.
The drives are 1tb seagate 7200.12 drives linked via 8088 cable to the server HBA card LSI 9200
 
rlksun@proxima:~# uname -a
SunOS proxima 5.11 oi_148 i86pc i386 i86pc
 
Because ZFS does error parity checking on reads, to compare the data read to be consistent with expected result from the parity data read. But the writes are "flat", basically, it does more read operations then write operations. Also, some read on write operations are probably done to confirm correct writes.
 
12 drives total in sas expander box connected to server via single 8088 cable
I know there can be many variables so the question isn't simple to answer but I was just suspecting that maybe block size or some other tuning parameter may need adjusting. Maybe next step is a benchmark test. If someone could suggest a good one and the approriate parameters I can run that and post the results
Thanks
 
HDDs write a bit faster than they read, so read speeds are generally slower than writes. That's going to be magnified by the raidz to some extent because it has the random read iops of a single drive but the sequential write iops of 10 drives. Because zfs uses copy on write, most writes are sequential writes, while reads aren't necessarily sequential.

AFAIK you will never see a raidz with faster raw read than write performance.
 
Last edited:
this is not true for large sequential reads. i routinely get 25% or more fast reads than writes. possibly as you say the random workload is different than sequential?
 
this is not true for large sequential reads. i routinely get 25% or more fast reads than writes.
Click to expand...
Is that the overall iostat output or the timed benchmark for a single operation? Because the latter will be influenced by arc/prefetching/atime, correct?
 
total time. the arc is not in play here, since we are talking about a single large file, well larger than the size of RAM, and it starts out cold anyway...
 
You must log in or register to reply here.
Back
Top