SSD Server Benchmarks
We thought we would publish some benchmarks of our latest generation of 3dpixel.net servers for both personal and dedicated customers.
Our primary aims are to:
a) increase performance
b) reduce power consumption
To this end, we are gradually replacing our Dual Core Nocona Xeon Servers with SCSI drives to Quad Core Harpertown Xeon Servers with the latest SSD drives from Intel.
Web serving in the main is an IO limited enterprise. This means that although CPU and memory speed are reasonably important, the speed of the hard drive subsystem is critical. This means that the drives themselves, and the controller used to power them are potentially the most important parts of any webserver.
We have opted for the new Intel X25 Solid State Drives, with all the latest firmwares applied combined with a 3Ware 9690SA RAID controller for RAID5 operations, which we have found to be the most suitable option for webserving (best reads).
Write test:
One of the supposed downsides of SSD drives is the write speed, perhaps moreso with RAID5 which has a poor write operation max compared to other RAID levels due to the complex parity calculations.
Creation of a 10GB file-
dd if=/dev/zero of=file_10GB bs=10000000 count=1k
1024+0 records in
1024+0 records out
10240000000 bytes (10 GB) copied, 61.8966 seconds, 165 MB/s
Blows the water out of our SCSI machines which usually write at about 120MB/s.
Read Test:
hdparm -t /dev/sda1
/dev/sda1:
Timing buffered disk reads: 1224 MB in 3.00 seconds = 408.85 MB/sec
Again, SCSI 15k in RAID5 reads are about 190MB/s albeit on a 3year old server.
The main point we have noticed is how fast the array seems to perform. Obviously the latency on the SSD drives is virtually non-existant versus ‘traditional media’. This should improve the general responsiveness of sites hosted.
Power Consumption:
CPUs aside, which use around 45w versus the 75w of the older Nocona Xeons (which saves 0.25amps by itself) we have worked out that saving the power of 4 15krpm SCSI drives compared to the virtual non-existant power consuming SSDs is about 50w which is 0.2amps. This means we have essentially reduced the power consumption of a standard shared hosting platform from 1amp to 0.6amps. Scaled up to a lot of servers, this is a massive power saving albeit at a cost. Having said that, the cost per GB is what we were paying 3/4 years ago for SCSI.
Green computing, with unbeatable performance 