You are here
SSD VPS performance chart
Important: this page is no longer updated. All the hosting providers/plans data are now created as independent entities.
All the actual entries are now added to a new SSD VPS performance chart (with sortable representation). When all the entries from this page are transferred, the mentioned link will become the only chart available. You will be able to comment on individual plans and hosting providers pages.
SSD (solid state drives) are becoming more popular, as their lifetime limits (maximal cell write cycles number) are growing. There are quite few hosting providers offering SSD as primary/aditional disk drives for their customers; since the I/O performance of SSD drives is typically 2-3 orders of magnitude higher than that of traditional magnetic drives, SSD-based plans could be used for disk-intensive tasks, such as Web and other applications heavily using databases.
Below is a performance chart I started to compose for VPS/dedicated servers/cloud hosting utilizing SSD as primary/auxiliary storage option. If the chart grows too large, it will be spread over several pages.
Note that I order the plans myself, so the speed of testing new providers is only limited by my monthly testing budget. Many thanks to those providers who offered their VPSes/cloud servers free of charge, for testing purposes.
Three primary results are given for the tested hosting plans: UnixBench, sequential write speed of a 1Gb file and random read/random write IOPS ratings. Scroll down to find what software is used and what commands were given to collect the information.
Note: this page will be constantly updated, visit it at a later time to see more entries/comments. See also the original thread on WHT.
Note: the tests won't give the same results if run again. That's normal. I ran several tests in different times of day and selected those most average. You can supply your version of tests to include in the table below. Thanks in advance.
Note: on a crowded host machine, as more customers get their VPS, the performance will inevitably deteriorate, simply because the same disk operations are shared among many users. Dedicated server and, at certain types of cloud hosting could provide you more or less constant performance.
Note: I recommend also visiting alternate benchmarking-related sites, such as ServerBear's Web Server Performance Leaderboard.
SSD VPS/DS/cloud plans benchmarks
The results are given in alphabetical order (sorted by provider's domain name). Wherever possible, hardware specs are given. zip link is a link to the archive with all the results (text files) included. Download them to gain all the info.
|Plan / price / technology||Datacenters||SSD model||CPU||RAM||UB||DD||IOPS|
|USA (Chicago, Las Vegas)||Samsung 830s (cache)||6x Xeon L5420 2.5Ghz||6Gb||2346.8||782||5236||7748|
VMWare VSphere zip
|OCZ Vertex 3||i7 2600 3.40Ghz||4Gb||1678.7||352||32084||39633|
Silver (standard performance)
|Crucial M4's or Samsung 830 series||Xeon E31230 3.20GHz||2Gb||696.9||274||21248||16353|
|Blue Light Host|
XEN/KVM mix zip
USA (East Coast)
|N/A(**)||4x Xeon E5645 2.40GHz||4Gb||1147.0||107||23019||8108|
70 USD/mo (*)
|Intel 320s and Samsung 830s||4x Xeon E31270 3.40GHz||1.5Gb||3908.8||376||113627||55454|
no minimal price, see here
USA (Las Vegas)
|N/A||1x AMD Opteron 2427 2.20GHz||2.0Gb||750.8||295||23591||10053|
SSD DotBlock 1
|QEMU Virtual CPU 3.00GHz||3Gb||699.7||28||3074||932|
none (dedicated server) zip
|France (Paris)||Adata S396||Atom D410 1.66GHz||1Gb||288.0||99||3986||9077|
|USA (Coloinseattle Tukwila)||Intel||QEMU virtual CPU 3.1GHz x 4||1.5Gb||3050.6||1400||135536||135501|
VPS 1GBit NL 2
|Rotterdam (Netherlands)||Crucial M4's (512GB) and Samsung 830 SSDs||2x Xeon E31230 3.20GHz||1Gb + 1Gb||1211.7||501||29825||34777|
|Russia (Moscow)||i7 3820 3.60GHz||1Gb||690.0||153||43448||11327|
Xpress RS SSD
|OCZ Vortex III SATA III||2x i7 2600 3.40GHz||1Gb+0.5Gb||1224.0||501||41228||19625|
|UK, Kent (Custodian DC)||N/A||Intel Xeon E3 3GHz||0.25Gb||1532.1||393||3716 (***) (59456)||1778 (28448)|
|UK||Intel 320s||Xeon W3520 2.67GHz||0.5Gb+0.5Gb||1094.5||346||42163||21769|
|USA (Kansas City)||Kingston SSDs||4x AMD FX(tm)-4100 3.60GHz||1.0Gb+1.0Gb||1111.5||217||25272||2640|
|Italy (Milan)||Intel 520 SSDSC2CW24||2x Xeon E31270 3.4GHz||1.5Gb||2372.2||496||47757||17433|
|USA (Dallas)||Intel 320 models||2x Xeon E31230 3.20GHz||0.5Gb||2448.4||289||90589||29475|
|USA (Dallas)||Samsung 830 (256Gb each)||2x Xeon E3-1230v2 3.30GHz||0.5Gb||2461.5||519||97243||69000|
none (dedicated server) zip
|USA (Dallas, SoftLayer DAL01)||Micron P300||8 x Xeon E3-1270 3.4GHz||32Gb||6316.5||1945||36488 (****)||21528|
|USA (Los Angeles, MultaCON datacenter)||4 x Xeon E5645 2.4GHz||0.25Gb / 0.5Gb||1162.1||367||50510||19037|
|USA (Chicago, Las Vegas)||Samsung 830s (cache)||2x Xeon L5420 2.5Ghz||1Gb||1187.4||754||5782||7441|
|USA (Chicago)||4x i3 540 3.07GHz||1Gb+1Gb||2586.6||165||34310||14415|
|Germany (Hetzner DC)||OCZ Vertex-3||Intel i7 2600 3.4GHz||0.25Gb + 0.125Gb||1239.2||238||26047||40226|
|France (OVH)||Intel 320s (300 Gb)||Xeon W3530 2.80GHz||1.0Gb+0.25Gb||359.2||99||13586||11429|
*: note that CleverKite's least expensive plan, worth 35 USD/mo, would give basically the same ratings.
**: it means the hosting provider prefers not to disclose their hardware components details.
***: IOPS capacity id distributed across all co-hosted VPSes. Thus, since VPS1 is hosted in amounts 16 per server, the overall capacity should be multiplied by 16 (values in parentheses).
****: the provider's management gave 170K random read IOPS. Unfortunately, the below tool gave lesser value. There are many factors that could influence the results (note that I, for example, flush caches before running fio one more time).
*****: this alternative configuration uses the following servers components: E3-1230v2, 4x256GB Samsung 830 SSD32GB RAM.
******: SSD-cached configurations.
Important: number of CPUs ('virtual cores') reported by UnixBench is usually taken from /proc/cpuinfo, or similar facility and may differ from actual CPU capability of the host computer.
For the UB (UnixBench) rating, Byte UnixBench was used. Standard 'make', './Run' commands were used to generate the results available.
For the I/O throughput measurement, Iozone Filesystem Benchmark is used. Although its results aren't shown in the table above, you can download .ZIP file containing all the used testing data, Iozone included. The following commands were used for every VPS/DS tested:
./iozone -O -R -a | tee -a report/auto.out ./iozone -O -R -r 4k -s 1m | tee -a report/1m.out ./iozone -O -R -r 4k -s 10m | tee -a report/10m.out ./iozone -O -R -r 4k -s 100m | tee -a report/100m.out ./iozone -O -R -r 4k -s 1000m | tee -a report/1000m.out
To gain IOPS random read/write ratings, fio tool was used. A test involving 32 processes, reading/writing out of/to 128Mb files in blocks of 4k during 10 seconds. The commands used to test are:
# wait 1 to 15 seconds sync ; echo 3 > /proc/sys/vm/drop_caches fio --bs=4k --size=128m --direct=1 --runtime=10 --rw=randread \ --numjobs=32 --group_reporting --time_based --name=128m | \ tee -a iops-32jobs-128m-4k-10sec-randread.txt # wait 1 to 15 seconds sync ; echo 3 > /proc/sys/vm/drop_caches fio --bs=4k --size=128m --direct=1 --runtime=10 --rw=randwrite \ --numjobs=32 --group_reporting --time_based --name=128m | \ tee -a iops-32jobs-128m-4k-10sec-randwrite.txt
To obtain DD, sequential writing speed (in megabytes per second, MB/s), the following command was used:
dd if=/dev/zero of=/home/2gb.dat bs=8k count=256k conv=fdatasync
The following hosting providers are in the queue at the moment (I will order a VPS from them and run tests as soon as possible, '*' means in process):
- (*) alpha1servers.com
- (*) burst.net
- (*) digitalocean.com
- (*) gigenetcloud.com
- (*) stormvz.com
If you could suggest more providers, please contact me.
I greatly appreciate the offers by Avante Hosting, CleverKite, GridVirt, VirtualSRV, KillerIO, Prometeus, SSD Nodes, HostHatch Inc, RockMyWeb DBA (on both its services, 50tbcloud.com and VirtuallyDedicated.com), Clustered.net, Castleweb Ltd, 24khost.com, WSWD.net, sugarhosts.com, ThePrimeHost and Fusioned.net representatives who kindly provided me with test VPSes and/or dedicated servers, respectively.
I greatly appreciate assistance of doughnet @ WHT having sent me the benchmark data on Pieserv SSD VPS mentioned above.