![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
After recent system upgrade, I now have a bit of time/resources to revisit about items I mentioned last time.
Original focus : Effect of Hybrid RAID-1 for Server with 2-disks (SSD + mechanical disk)
Status : I could not afford a full-scale all physical setup to test the hypothesis.
Counter-measure : So I look for the next best thing by doing the following.
A. I am going to simulate the effect. For various reasons it is done inside a VM.
1. Setup VM-1 with 1 OS disk and 1-CPU
2. Setup 1GB RAMDISK inside VM-1 --> /dev/ram1
3. Setup 4GB additional virtual disk --> /dev/sdb
Testing Results
1. seeker one thread against /dev/ram1 --> 175946 seeks/second 0.006 ms (Ramdisk)
2. seeker one thread against /dev/sdb --> 15488 seeks/second 0.065 ms (VM cache I/O)
3. setup mdadm RAID-1 /dev/md1 with /dev/ram1 and /dev/sdb as members
3a. seeker one thread against /dev/md1 --> 74577 seeks/second 0.013 ms
4. setup mdadm RAID-1 /dev/md1 with /dev/ram1 and /dev/sdb as members. /dev/sdb as write_mostly
4a. seeker one thread against /dev/md1 --> 154742 seeks/second 0.006 ms
So for seeker (Random read test), results in point-1 and point-4 differ slightly only
Note: the value is very fast due to cache I/O-VM-1 and Ramdisk in use.
Tentative : as we can observe, even cache I/O is still slower than RAM disk, approximately 10-times. Anyhow For mostly READ-heavy setup, this is worth investigation.
Now
1. replace RAM disk with SSD
2. replace cache I/O with mechanical SATA disk.
the value/ratio likely different, but the general hypothesis should still valid in this case. Obviously seeker is not a conclusive test. Maybe other more experience users can give better tests/benchmarks to real-world conditions.
Original focus : Effect of Hybrid RAID-1 for Server with 2-disks (SSD + mechanical disk)
Status : I could not afford a full-scale all physical setup to test the hypothesis.
Counter-measure : So I look for the next best thing by doing the following.
A. I am going to simulate the effect. For various reasons it is done inside a VM.
1. Setup VM-1 with 1 OS disk and 1-CPU
2. Setup 1GB RAMDISK inside VM-1 --> /dev/ram1
3. Setup 4GB additional virtual disk --> /dev/sdb
Testing Results
1. seeker one thread against /dev/ram1 --> 175946 seeks/second 0.006 ms (Ramdisk)
2. seeker one thread against /dev/sdb --> 15488 seeks/second 0.065 ms (VM cache I/O)
3. setup mdadm RAID-1 /dev/md1 with /dev/ram1 and /dev/sdb as members
3a. seeker one thread against /dev/md1 --> 74577 seeks/second 0.013 ms
4. setup mdadm RAID-1 /dev/md1 with /dev/ram1 and /dev/sdb as members. /dev/sdb as write_mostly
4a. seeker one thread against /dev/md1 --> 154742 seeks/second 0.006 ms
So for seeker (Random read test), results in point-1 and point-4 differ slightly only
Note: the value is very fast due to cache I/O-VM-1 and Ramdisk in use.
Tentative : as we can observe, even cache I/O is still slower than RAM disk, approximately 10-times. Anyhow For mostly READ-heavy setup, this is worth investigation.
Now
1. replace RAM disk with SSD
2. replace cache I/O with mechanical SATA disk.
the value/ratio likely different, but the general hypothesis should still valid in this case. Obviously seeker is not a conclusive test. Maybe other more experience users can give better tests/benchmarks to real-world conditions.