Untuned Ceph Setup quite slow
Pages: 1 2
fx882
17 Posts
September 22, 2017, 1:13 pmQuote from fx882 on September 22, 2017, 1:13 pmOk. Surprise Surprise. I got VMWare up and running. The installed VM(Debian 9.0) has this test results:
root@ceph-vm2:~# timeout 900 ./script.sh ; timeout 900 ./script-read.sh
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.04998 s, 173 MB/s in 0m6.052s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.07296 s, 173 MB/s in 0m6.074s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.49803 s, 191 MB/s in 0m5.500s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.50457 s, 233 MB/s in 0m4.506s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.21915 s, 249 MB/s in 0m4.221s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.10759 s, 255 MB/s in 0m4.109s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.2695 s, 246 MB/s in 0m4.271s
root@ceph-vm2:~# timeout 900 ./script-read.sh
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.81191 s, 154 MB/s in 0m6.813s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.09934 s, 206 MB/s in 0m5.101s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.37752 s, 195 MB/s in 0m5.379s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.21591 s, 201 MB/s in 0m5.217s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.46832 s, 192 MB/s in 0m5.470s
That's definitively something usable.
EDIT-1
But I do not see any Traffic on iSCSI-1/iSCSI-2 Network at all. As if VMWare does some very strange caching here.
EDIT-2
Ok. One watching terminal was wrong. There is indeed iSCSI-traffic. But it seems VMWare does extensive Caching. So 1 GB data did not get through to the storage. I rewrote my testing script and writing continuously high amounts of random data. Even then VMWare is about 2x faster than citrix. The network saturation is a lot higher in VMWare than in Citrix. Since the test systems are not completely identical it may be differences in hardware.
Ok. Surprise Surprise. I got VMWare up and running. The installed VM(Debian 9.0) has this test results:
root@ceph-vm2:~# timeout 900 ./script.sh ; timeout 900 ./script-read.sh
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.04998 s, 173 MB/s in 0m6.052s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.07296 s, 173 MB/s in 0m6.074s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.49803 s, 191 MB/s in 0m5.500s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.50457 s, 233 MB/s in 0m4.506s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.21915 s, 249 MB/s in 0m4.221s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.10759 s, 255 MB/s in 0m4.109s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.2695 s, 246 MB/s in 0m4.271s
root@ceph-vm2:~# timeout 900 ./script-read.sh
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.81191 s, 154 MB/s in 0m6.813s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.09934 s, 206 MB/s in 0m5.101s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.37752 s, 195 MB/s in 0m5.379s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.21591 s, 201 MB/s in 0m5.217s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.46832 s, 192 MB/s in 0m5.470s
That's definitively something usable.
EDIT-1
But I do not see any Traffic on iSCSI-1/iSCSI-2 Network at all. As if VMWare does some very strange caching here.
EDIT-2
Ok. One watching terminal was wrong. There is indeed iSCSI-traffic. But it seems VMWare does extensive Caching. So 1 GB data did not get through to the storage. I rewrote my testing script and writing continuously high amounts of random data. Even then VMWare is about 2x faster than citrix. The network saturation is a lot higher in VMWare than in Citrix. Since the test systems are not completely identical it may be differences in hardware.
Last edited on September 22, 2017, 3:01 pm by fx882 · #11
Pages: 1 2
Untuned Ceph Setup quite slow
fx882
17 Posts
Quote from fx882 on September 22, 2017, 1:13 pmOk. Surprise Surprise. I got VMWare up and running. The installed VM(Debian 9.0) has this test results:
root@ceph-vm2:~# timeout 900 ./script.sh ; timeout 900 ./script-read.sh
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.04998 s, 173 MB/s in 0m6.052s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.07296 s, 173 MB/s in 0m6.074s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.49803 s, 191 MB/s in 0m5.500s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.50457 s, 233 MB/s in 0m4.506s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.21915 s, 249 MB/s in 0m4.221s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.10759 s, 255 MB/s in 0m4.109s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.2695 s, 246 MB/s in 0m4.271s
root@ceph-vm2:~# timeout 900 ./script-read.sh
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.81191 s, 154 MB/s in 0m6.813s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.09934 s, 206 MB/s in 0m5.101s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.37752 s, 195 MB/s in 0m5.379s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.21591 s, 201 MB/s in 0m5.217s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.46832 s, 192 MB/s in 0m5.470sThat's definitively something usable.
EDIT-1
But I do not see any Traffic on iSCSI-1/iSCSI-2 Network at all. As if VMWare does some very strange caching here.
EDIT-2
Ok. One watching terminal was wrong. There is indeed iSCSI-traffic. But it seems VMWare does extensive Caching. So 1 GB data did not get through to the storage. I rewrote my testing script and writing continuously high amounts of random data. Even then VMWare is about 2x faster than citrix. The network saturation is a lot higher in VMWare than in Citrix. Since the test systems are not completely identical it may be differences in hardware.
Ok. Surprise Surprise. I got VMWare up and running. The installed VM(Debian 9.0) has this test results:
root@ceph-vm2:~# timeout 900 ./script.sh ; timeout 900 ./script-read.sh
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.04998 s, 173 MB/s in 0m6.052s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.07296 s, 173 MB/s in 0m6.074s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.49803 s, 191 MB/s in 0m5.500s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.50457 s, 233 MB/s in 0m4.506s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.21915 s, 249 MB/s in 0m4.221s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.10759 s, 255 MB/s in 0m4.109s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 4.2695 s, 246 MB/s in 0m4.271s
root@ceph-vm2:~# timeout 900 ./script-read.sh
1048576000 bytes (1.0 GB, 1000 MiB) copied, 6.81191 s, 154 MB/s in 0m6.813s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.09934 s, 206 MB/s in 0m5.101s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.37752 s, 195 MB/s in 0m5.379s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.21591 s, 201 MB/s in 0m5.217s
1048576000 bytes (1.0 GB, 1000 MiB) copied, 5.46832 s, 192 MB/s in 0m5.470s
That's definitively something usable.
EDIT-1
But I do not see any Traffic on iSCSI-1/iSCSI-2 Network at all. As if VMWare does some very strange caching here.
EDIT-2
Ok. One watching terminal was wrong. There is indeed iSCSI-traffic. But it seems VMWare does extensive Caching. So 1 GB data did not get through to the storage. I rewrote my testing script and writing continuously high amounts of random data. Even then VMWare is about 2x faster than citrix. The network saturation is a lot higher in VMWare than in Citrix. Since the test systems are not completely identical it may be differences in hardware.