[DRBD-user] Suspicious Performance Gain

[Michael Hierweck]

Hi all,

we use hardware Raid controllers with BBU and 2.5" SAS3 drives in RAID 10. On top we use LVM
fat provisioning. DRBD is 8.4.

1) HWR10 => LVM => XFS
2) HWR10 => LVM => DRBD "C" => QEMU/Virtio => XFS

DRBD settings:
	disk-flushes = no
	md-flushes = no

Virtio settings:
	cache = none

Performing benchmarks with fio (direct=1, sync=1, bs=1k,2k,4k,...,128k) leads to suspicious.

While 1) leads to expected results such as 200-400 IOPS on random writes (expected value) 2)
leads to about 4 times better results.

(I did the same benchmarks some years ago on the same hardware but with an older software stack
and remember that DRBD protocol "C" lead to a performance decrease of about factor 2. Factor 2
seems to be a good value for synchronous replication.)

What's going wrong here?

Thanks in advance,

Michael