So we were discussing SDD versus HDD, and it made me think: While SDD is faster than HDD; how much faster is it in reality, i.e. what impact does it have on general use? So I tested it:
Test setup:
1 x 2GB RAMDISK (I do not have an SDD)
1 x 300GB Seagate HDD
I created a single Windows XP SP3 VM on a 1.4 GB vdisk, running VMware workstation 6.5.
Installed HDTune and vmtools. This vdisk was then copied to a 2GB RAM disk, and a 300 GB HDD.
I then created a batch file to run in the startup folder, to time-stamp a point in the Windows startup. I used the VMlog to measure the exact time the VM started (pushed the play button), i.e. to get the beginning of the startup routine and a point near the end of the startup routine.
Benchmark
HDTune of vdisk on the 300GB HDD (Average 3 runs)
Min 28.0 MB/s
Max 82.7 MB/s
Average 76.2 MB/s
Access Time 6.5 ms
Burst 718.2 MB/s
HDTune of vdisk on 2GB RAMDISK (Average of 3 runs)
Min 259.8 MB/s
Max 509.0 MB/s
Average 464.4 MB/s
Access Time 0.1 ms
Burst 564.3 MB/s
Startup of Windows on vdisk on the 300GB HDD (Average 3 runs)
42 Seconds
Startup of Windows on vdisk on 2GB RAMDISK (Average of 3 runs)
35 Seconds
Conclusion
A Ramdisk, and therefore an SDD, blows the pants off an HDD in a benchmark. However, in a realistic measurement (Windows startup), it would seem that the impressive benchmark translates to a slightly less dramatic improvement in startup times.
A quick review of the performance monitor (running a VM allows one to measure the underlying hardware at startup, which cannot be done on a physical machine), suggests a performance bottleneck at the processor, as well as a lower than anticipated IO rate on the physical disk.
This reliance on the CPU, and lower than expected disk IO, further reduces the positive impact of a faster disk. If the disk is only used for 20% of the startup routine, then it's impact would be minimal.
Of course this is only a single test and other applications may perform differently, especially those with a high disk IO, low CPU rate. I must mention that the XP install took less than 7 minutes, from start to finish (automated XP install, including VMtools install
), and in that instance the Ramdisk was REALLY quick (during the file copy part).
It should also be noted that an SDD may have a reduced performance when compared to a RAMdisk, as the SATA interface is slower than RAM. So these figures represent the upper mark for current and short term SDD disks.
Unfortunately SDD disks are still too small, and therefore not suitable for storage, and will therefore not fully benefit from the improved speed....
Test setup:
1 x 2GB RAMDISK (I do not have an SDD)
1 x 300GB Seagate HDD
I created a single Windows XP SP3 VM on a 1.4 GB vdisk, running VMware workstation 6.5.
Installed HDTune and vmtools. This vdisk was then copied to a 2GB RAM disk, and a 300 GB HDD.
I then created a batch file to run in the startup folder, to time-stamp a point in the Windows startup. I used the VMlog to measure the exact time the VM started (pushed the play button), i.e. to get the beginning of the startup routine and a point near the end of the startup routine.
Benchmark
HDTune of vdisk on the 300GB HDD (Average 3 runs)
Min 28.0 MB/s
Max 82.7 MB/s
Average 76.2 MB/s
Access Time 6.5 ms
Burst 718.2 MB/s
HDTune of vdisk on 2GB RAMDISK (Average of 3 runs)
Min 259.8 MB/s
Max 509.0 MB/s
Average 464.4 MB/s
Access Time 0.1 ms
Burst 564.3 MB/s
Startup of Windows on vdisk on the 300GB HDD (Average 3 runs)
42 Seconds
Startup of Windows on vdisk on 2GB RAMDISK (Average of 3 runs)
35 Seconds
Conclusion
A Ramdisk, and therefore an SDD, blows the pants off an HDD in a benchmark. However, in a realistic measurement (Windows startup), it would seem that the impressive benchmark translates to a slightly less dramatic improvement in startup times.
A quick review of the performance monitor (running a VM allows one to measure the underlying hardware at startup, which cannot be done on a physical machine), suggests a performance bottleneck at the processor, as well as a lower than anticipated IO rate on the physical disk.
This reliance on the CPU, and lower than expected disk IO, further reduces the positive impact of a faster disk. If the disk is only used for 20% of the startup routine, then it's impact would be minimal.
Of course this is only a single test and other applications may perform differently, especially those with a high disk IO, low CPU rate. I must mention that the XP install took less than 7 minutes, from start to finish (automated XP install, including VMtools install
), and in that instance the Ramdisk was REALLY quick (during the file copy part). It should also be noted that an SDD may have a reduced performance when compared to a RAMdisk, as the SATA interface is slower than RAM. So these figures represent the upper mark for current and short term SDD disks.
Unfortunately SDD disks are still too small, and therefore not suitable for storage, and will therefore not fully benefit from the improved speed....
