Hi,
Does anyone else notice the extreme slow guest OS startup under VirtualBox since OL 8? The startup of a VM running OL 8.1 takes about 5 times longer compared to OL 7.8.
I tested a couple of different systems. All virtual guest machines are minimal installations (no GUI) with the latest updates installed. They all use the same VM settings with 2GB of RAM, same disk type (Samsung SSD 850 1TB), same host OS (10.9.5). I also tried VirtualBox 5.1.38 and 6.1.6, but the performance differences between the two versions were only marginal. To avoid ill affects caused the host system, all tests were run after a fresh system restart with nothing else running. Following OS guest system startup times:
Hardware | OL 7.8 UEK 5 | OL 8.1 UEK 6 | OL 8.1 RHCK |
---|
Mac Mini 2012 16GB RAM 1TB SSD: # cat /proc/cpuinfo | grep model model : 58 model name : Intel(R) Core(TM) i5-3210M CPU @ 2.50GHz | 7s | 32s | 34s |
MacBook Pro 2009 8GB RAM 1TB SSD # cat /proc/cpuinfo | grep model model : 23 model name : Intel(R) Core(TM)2 Duo CPU P8700 @ 2.53GHz | 1m 2s | 1m 54s | 2m 2s |
All guest machines use the same drivers and same virtual hardware:
OL 7.8 UEK 5 | OL 8.1 UEK 6 |
---|
00:0d.0 SATA controller [0106]: Intel Corporation 82801HM/HEM (ICH8M/ICH8M-E) SATA Controller [AHCI mode] [8086:2829] (rev 02) Kernel driver in use: ahci Kernel modules: ahci # modinfo ahci filename: /lib/modules/4.14.35-1902.301.1.el7uek.x86_64/kernel/drivers/ata/ahci.ko.xz version: 3.0 | 00:0d.0 SATA controller [0106]: Intel Corporation 82801HM/HEM (ICH8M/ICH8M-E) SATA Controller [AHCI mode] [8086:2829] (rev 02) Kernel driver in use: ahci Kernel modules: ahci # modinfo ahci filename: /lib/modules/5.4.17-2011.1.2.el8uek.x86_64/kernel/drivers/ata/ahci.ko.xz version: 3.0 |
Below are test results of the Mac Mini 2012 and MacBook Pro late 2009. The tests were run several times to get some average values.
OL 7.8 UEK 5 | OL 8.1 UEK 6 |
---|
# hdparm -tT /dev/sda /dev/sda: Timing cached reads: 11064 MB in 1.99 seconds = 5547.55 MB/sec Timing buffered disk reads: 2842 MB in 3.00 seconds = 946.83 MB/sec | # hdparm -tT /dev/sda /dev/sda: Timing cached reads: 7590 MB in 1.99 seconds = 3807.02 MB/sec Timing buffered disk reads: 126 MB in 3.09 seconds = 40.81 MB/sec |
# cryptsetup benchmark -c aes-xts-plain64 -s 512 # Tests are approximate using memory only (no storage IO). # Algorithm | Key | Encryption | Decryption aes-xts 512b 808.4 MiB/s 818.1 MiB/s | # cryptsetup benchmark -c aes-xts-plain64 -s 512 # Tests are approximate using memory only (no storage IO). # Algorithm | Key | Encryption | Decryption aes-xts 512b 802.3 MiB/s 798.8 MiB/s |
Below are test results on the MacBook Pro late 2009:
OL 7.8 UEK 5 | OL 8.1 UEK 6 |
---|
# hdparm -tT /dev/sda /dev/sda: Timing cached reads: 2700 MB in 1.97 seconds = 1370.28 MB/sec Timing buffered disk reads: 838 MB in 3.00 seconds = 279.21 MB/sec | # hdparm -tT /dev/sda /dev/sda: Timing cached reads: 2506 MB in 1.97 seconds = 1272.58 MB/sec Timing buffered disk reads: 682 MB in 3.01 seconds = 226.88 MB/sec |
# cryptsetup benchmark -c aes-xts-plain64 -s 512 # Tests are approximate using memory only (no storage IO). # Algorithm | Key | Encryption | Decryption aes-xts 512b 75.6 MiB/s 90.2 MiB/s | # cryptsetup benchmark -c aes-xts-plain64 -s 512 # Tests are approximate using memory only (no storage IO). # Algorithm | Key | Encryption | Decryption aes-xts 512b 79.8 MiB/s 89.1 MiB/s |
On the Mac Mini 2012, disk performance under OL 8.1 drops down to 41 MB/s compared to OL 7.8 at 947 MB/s.
This may explain the slow startup between the two systems, i.e. 7s vs. 32s, but not really.
Considering that the Mac Mini supports SATA3 and the SSD supports it too, disk performance should be better compared to the MacBook Pro which only supports SATA2. However, disk performance on the MacBook Pro with OL 7 and OL 8 are both much better with OL 7 at 272 MB/s..
This does not explain the slow startup, i.e. 1m 02s vs. 1m 52s, nor does it explain the difference running on a different CPU.
The MacBook Pro CPU does not have AES instructions, hence the different encoding performance. Why timing cached reads are so much slower compared to the Mac Mini is questionable, but the MacBook Pro does use slower RAM: DDR3 1066 MHz vs 1600 MHz.
OL 8 beta did not support the Intel Dual Core processor and also RHEL 8 required a kernel update in the GA release. But the slow disk performance of OL 8 using an Intel Core i5 looks very strange to me. As I recall, OS release 8 supports the Intel iCore CPU.
I guess I could use dd to test various options, but at this point the problem seems CPU dependent.
Any ideas? Thank you.