Post by tomharvey on Nov 30, 2009 22:59:13 GMT 7
We are using a N8800 and despite good write performance when writing to the RAID, we are seeing terrible read performance if we want to re-access this data. This is not an issue when accessing the iSCSI partitions of the system, but only when accessing the AFP share. Is this a known issue with AFP (I believe they are using Netatalk) and would be best off accessing this through an NFS share?
Curiously this only seems to be the case when accessing small blocks - see the benchmarks below. So is this an issue with the network setup; I have jumbo frames on and is supported by our router.
The RAID reports as being fully healthy, using all 8 disks and only 4.8TB of the 13TB capacity is used and the disk is formatted for XFS. This issue is seen when only one machine is wired to the RAID, so along with the good iSCSI performance this does not seem to be a network traffic issue.
Thanks for your suggestions, benchmarks below:
AFP share
Sequential
Uncached Write 12.73 MB/sec [4K blocks]
Uncached Write 47.62 MB/sec [256K blocks]
Uncached Read 12.55 MB/sec [4K blocks]
Uncached Read 2.16 MB/sec [256K blocks]
Random
Uncached Write 12.40 MB/sec [4K blocks]
Uncached Write 48.81 MB/sec [256K blocks]
Uncached Read 11.80 MB/sec [4K blocks]
Uncached Read 2.23 MB/sec [256K blocks]
iSCSI Target
Sequential
Uncached Write 49.58 MB/sec [4K blocks]
Uncached Write 43.07 MB/sec [256K blocks]
Uncached Read 11.01 MB/sec [4K blocks]
Uncached Read 8.80MB/sec [256K blocks]
Random
Uncached Write 42.37 MB/sec [4K blocks]
Uncached Write 45.74 MB/sec [256K blocks]
Uncached Read 10.69 MB/sec [4K blocks]
Uncached Read 11.45 MB/sec [256K blocks]
Curiously this only seems to be the case when accessing small blocks - see the benchmarks below. So is this an issue with the network setup; I have jumbo frames on and is supported by our router.
The RAID reports as being fully healthy, using all 8 disks and only 4.8TB of the 13TB capacity is used and the disk is formatted for XFS. This issue is seen when only one machine is wired to the RAID, so along with the good iSCSI performance this does not seem to be a network traffic issue.
Thanks for your suggestions, benchmarks below:
AFP share
Sequential
Uncached Write 12.73 MB/sec [4K blocks]
Uncached Write 47.62 MB/sec [256K blocks]
Uncached Read 12.55 MB/sec [4K blocks]
Uncached Read 2.16 MB/sec [256K blocks]
Random
Uncached Write 12.40 MB/sec [4K blocks]
Uncached Write 48.81 MB/sec [256K blocks]
Uncached Read 11.80 MB/sec [4K blocks]
Uncached Read 2.23 MB/sec [256K blocks]
iSCSI Target
Sequential
Uncached Write 49.58 MB/sec [4K blocks]
Uncached Write 43.07 MB/sec [256K blocks]
Uncached Read 11.01 MB/sec [4K blocks]
Uncached Read 8.80MB/sec [256K blocks]
Random
Uncached Write 42.37 MB/sec [4K blocks]
Uncached Write 45.74 MB/sec [256K blocks]
Uncached Read 10.69 MB/sec [4K blocks]
Uncached Read 11.45 MB/sec [256K blocks]