Dediserve DigitalOcean Linode Vultr Benchmarks DigitalOcean vs Linode vs Vultr vs Dediserve - London vs Singapore VPS benchmarks

Below are VPS benchmarks for DigitalOcean versus Linode versus Vultr versus Dediserve for the geographical regions for London and Singapore only. Vultr VPS benchmarks are only for London seeing as they do not have a Singapore location available at the time of this testing. August 24, 2015 update: Seems i mixed up the numbers for Linode Singapore E5-2680v3 KVM column was switched with Linode London E5-2680v2 KVM so bare that in mind

Benchmarks were done as my new centminmod.com site is deployed across 13+ VPS servers in various geographical locations and Singapore and London locations are apart of the mix. I need to find out exactly which is the best VPS provider for London and Singapore that is under the budget of ~US$40/month.

You can check out other comparison benchmarks here.

---

VPS servers:

All 7x VPS servers are installed with my own Centmin Mod 1.2.3.eva2000.08 stable LEMP web stack environment consisting of:

• CentOS 6.7 64 bit
• Nginx 1.9.3 with stream, SPDY/3.1, Nginx lua and ngx_pagespeed module support
• PHP 5.4.44 (PHP-FPM) with Zend Opcache 7.0.6
• MariaDB 10.0.21 MySQL
• Memcached 1.4.24 with memcache/memcached/libmemcached support
• CSF Firewall

Server Specs & Pricing

• DigitalOcean KVM VPS - Singapore & London: 4GB, 2 CPUS, 60GB SSD 4TB (1Gbps network in/out) US$40/month
• Linode KVM VPS - Singapore & London: 4GB, 4 CPUS, 96GB SSD 4TB (40Gbps in & 500Mbps network out) US$40/month
• Dediserve XEN VPS - Singapore & London: 4GB 4 CPUS, 50GB SSD + 50GB SAN +2 ips with bandwidth quotas calculated by per GB of memory installed. So for 4GB of memory = 4x.1.5TB = 6TB for London and 4x1TB = 4TB for Singapore (500Mbit/s network in/out upgraded from normal 250Mbit/s). So bare in mind below bandwidth tests maybe 1/2 the speed for normal 250Mbit/s VPSes. Terms of service have different network bandwidth restrictions for Premium locations like Singapore and also may have other bandwidth throttling for unsupported bandwidth traffic patterns. Price is usually US$41.76/month however had taken advantage of their 50% off for life offer so priced at US$20.88/month
• Vultr KVM VPS - 2.40Ghz London: 4GB, 2 CPUS, 90GB SSD 4TB (1Gbps network in/out) US$40/month currently discounted to US$32/month

Testing Methodology:

I basically created each respective VPS server using CentOS 6.5/6.6 64bit OS and ran YUM update to bring them all inline with CentOS 6.7 64bit OS. Then ran the new installnbench.sh file at centminmod/centminmodbench · GitHub which I wrote specifically for this testing. It basically does 2 tasks when run installnbench.sh via curl command line on a fresh CentOS 6.x or 7.x OS server:

1. install latest Centmin Mod .08 stable LEMP stack ~ 15-30 mins time then
2. installs and runs centminmodbench.sh (UnixBench enabled) ~ 30-60 mins time

Code:

curl -sL https://github.com/centminmod/centminmodbench/raw/master/installnbench.sh | bash

Linux Kernel:

The list of Linux kernels used for the 7x VPS servers

Vultr VPS kernels

Code:

uname -r
2.6.32-504.30.3.el6.x86_64

DigitalOcean VPS kernel

Code:

uname -r
2.6.32-504.12.2.el6.x86_64

Linode KVM Kernel

Code:

uname -r
4.1.0-x86_64-linode59

Dediserve Xen Kernel

Code:

uname -r
2.6.32-504.12.2.el6.x86_64

VPS Intel CPU Clock Speed:

• Linode London with Intel Xeon E5-2680v2 at 2.80ghz
• Linode Singapore with Intel Xeon E5-2680v3 at 2.50Ghz
• DigitalOcean with Intel Xeon E5-2630Lv2 at 2.40Ghz
• Dediserve with Intel Xeon E5-2660v1 at 2.20Ghz
• Vultr with Intel cpu with masked indentification at 2.40Ghz

 

Centmin Mod .08 Stable LEMP Install Times:

Centmin Mod .08 stable LEMP web stack install times are highly variable as part of the time is downloading source tarballs or YUM package downloads and installs so depends on the speed of the respective YUM mirrors the system picks up to download from as well as location of the VPS server geographically to the downloads. Downloads account for between 15-33% of the total install time. The rest is source compilation for Nginx, PHP-FPM, PHP extensions, Memcached server, CSF Firewall and other installed items via the Centmin Mod auto installer. Ultimately, comes down to cpu processor, disk performance and cpu clock speed for source compilation parts.

For installnbench.sh users, you'll find the install time on the very last line in the /root/centminlogs/centminmod_1.2.3-eva2000.08_***-***_install.log install log.

Centmin Mod LEMP .08 stable LEMP stack install has 2 parts, the YUM related download parts and the total install time which includes YUM, tarball download and source compilation. Centminmodbench.sh time is the total time it took to run the centminmodbench.sh script itself.

• Vultr is just a fraction of it's former self in terms of performance now that they have downgraded their cpus from 3.4+ Ghz to 2.4 Ghz and are using less performant disk configurations now. Almost twice as slow as 3.4+ Ghz instances for Centmin Mod install times.
• Linode has been very consistent over the months I have benchmarked them. Same can be said this time with great balance of network connectivity and cpu performance on both Linode London and Linode Singapore VPS locations.
• Dediserve is the newest in comparison benchmarks and I have benchmarked them a few times for Singapore and London locations and expected performance is about where it is given the slower processors used. Network connectivity is much slower on Dediserve locations and that is also reflected in the Centmin Mod install times. With that said, Dediserve London did rank 3rd for Centmin Mod install times. However, Dediserve Singapore was the slowest of all VPSes.
• DigitalOcean has gotten alot slower than in the past with disk I/O quite poor once you read further on down this thread for specific tests.

Centmin Mod is installed via one liner curl method which results in detailed breakdown of timed statistics at end of install.

Code:

curl -sL http://centminmod.com/installer.sh | bash

August 28, 2015 update: Seems i mixed up the numbers for Linode Singapore E5-2680v3 KVM column was switched with Linode London E5-2680v2 KVM so bare that in mind

 

centminmodbench.sh Results:

Ioping Disk and Ioping RamDisk memory test for random I/O performance. August 24, 2015 update: Seems i mixed up the numbers for Linode Singapore E5-2680v3 KVM column was switched with Linode London E5-2680v2 KVM so bare that in mind

In terms of IOPs default disk performance leader is Dediserve London and Dediserve Singapore. Followed by Linode Singapore and Linode London and then Vultr London, DigitalOcean London and last place with disk congestion and contention clearly showing is DigitalOcean Singapore.



RamDisk - testing system memory bandwidth

Clear leader is Linode Singapore and Linode London when it comes to system memory bandwidth performance. Followed by Dediserve London and Singapore, then Vultr London and with DigitalOcean having an unusually high default result but sequential and cached IOPs speed was very low.

 

DD Disk Benchmarks

August 24, 2015 update: Seems i mixed up the numbers for Linode Singapore E5-2680v3 KVM column was switched with Linode London E5-2680v2 KVM so bare that in mind

Disk

Linode London and Singapore are the clear winners here. Followed by Dediserve London and Singapore and then Vultr London. Clearly DigitalOcean London second last and DigitalOcean Singapore has serious disk performance issues which were also reflected in above disk performance tests coming dead last



Ramdisk

Linode London and Singapore have the higher memory bandwidth numbers. Followed by Dediserve. Vultr is seriously a fraction of it's former self, I'd say around 1/3 rd of the numbers compared to their pre-downgraded CPUs and thus dropping to DigitalOcean levels of system memory bandwidth.

 

Disk FIO benchmarks

August 24, 2015 update: Seems i mixed up the numbers for Linode Singapore E5-2680v3 KVM column was switched with Linode London E5-2680v2 KVM so bare that in mind

The best balance of random read and write bandwidth/IOPs goes to Linode London and Linode Singapore. Followed by Dediserve London, then Vultr and Dediserve Singapore. DigitalOcean Singapore while it has high random read, the random writes was very poor at 25MB/s respectively. DigitalOcean London also had poor random writes at 65MB/s

 

VPS Bandwidth Tests

Tested both single thread wget and multi-threaded parallel axel bandwidth tests. However, seems axel was broken and seems to give no results. So left it out of the tabulated results. Bear in mind geographical location of VPSes in relation to bandwidth numbers.

wget bandwidth

In terms of Cachefly bandwidth numbers which try to download from closest location to the VPS, Linode London and Singapore were clearly much faster thanks to Linode's fast network.

You can see clearly that Dediserve's network connectivity was restricted due to 500Mbit/s cap and also possibly due to Dediserve throttle bandwidth? I did have issues with Dediserve and CentOS 7 templates with kernel and bandwidth speed issues for London, Singapore and Sydney locations in the past. So I chose to use CentOS 6.7 for these tests. However, the result is pretty much the same Dediserve London and more so Dediserve Singapore bandwidth is much slower for locations outside of their VPS regions which can also be said for Vultr.

If local visitor speed is important, then for London it would be Linode > Vultr > DigitalOcean > Dediserve. For Singapore it would be Linode > DigitalOcean > Dediserve.

 

VPS Ping Tests

Ping tests are all geographically relative to the VPS server location.

 

VPS OpenSSL 1.01e system benchmarks

August 24, 2015 update: Seems i mixed up the numbers for Linode Singapore E5-2680v3 KVM column was switched with Linode London E5-2680v2 KVM so bare that in mind The Linode Singapore E5-2680v3 KVM seems to have aes cpu flag disabled so AES-NI hardware assistanced performance is halved compared to E5-2680v2 KVM which as the aes cpu flag.

Very clear that if you run a https SSL based site or are doing any type of OpenSSL encryption or decryption, that you do not want to be using Vultr 2.40Ghz or DigitalOcean VPS instances. Dediserve is only slightly better. But clear leader is Linode Singapore and Linode London.

Bear in mind though Linode and Dediserve also have more cpu cores at 4 which reflect in their results compared to Vultr and DigitalOcean only having 2 cpu cores.

VPS Nginx static LibreSSL 2.2.1 benchmarks

August 24, 2015 update: Seems i mixed up the numbers for Linode Singapore E5-2680v3 KVM column was switched with Linode London E5-2680v2 KVM so bare that in mind The Linode Singapore E5-2680v3 KVM seems to have aes cpu flag disabled so AES-NI hardware assistanced performance is halved compared to E5-2680v2 KVM which as the aes cpu flag.

Centmin Mod 1.2.3-eva2000.08's Nginx static compile switched from OpenSSL 1.0.2d to LibreSSL 2.2.1.

If you plan on using more performant based ECC 256 bit based SSL certificates using ECDSA then you need to look at Linode as they're using Intel Xeon E5-2680v2 cpus for London and Intel Xeon E5-2680v3 for Singapore. The same can be said for traditional RSA 2048 bit SSL certificates.

Linode Singapore has better hardware AES-NI assisted numbers for evp aes128/aes256 than Linode London due to the newer Intel Xeon E5-2680v3 being used compared to older generation Intel Xeon E5-2680v2

VPS Entropy Availability

Entropy availability pretty close seeing as Centmin Mod LEMP stack install by default tweaks and improves this thanks to using haveged.

 

mysqlslap MySQL Benchmarks

August 24, 2015 update: Seems i mixed up the numbers for Linode Singapore E5-2680v3 KVM column was switched with Linode London E5-2680v2 KVM so bare that in mind

mysqlslap tool is included with all MySQL installs including Oracle MySQL, MariaDB MySQL and Percona MySQL. The custom mysqlslap parameters I used are just arbitrarily in nature and only test MyISAM performamce across 64 clients.

Actual test command used is:

MariaDB 10.0.21 MySQL is the default installed MySQL version in Centmin Mod .08 stable LEMP auto installer.

mysqlslap results

Processor performance and clock speed plays a factor. MySQL slap though doesn't reflect real world usage so bear that in mind.

 

PHP benchmark

August 24, 2015 update: Seems i mixed up the numbers for Linode Singapore E5-2680v3 KVM column was switched with Linode London E5-2680v2 KVM so bare that in mind

• PHP benchmarks test the Centmin Mod .08 stable installed PHP 5.4.44 (php-fpm) binary (which defaults include enabling Zend Opcache 7.0.6) and tests against the PHP 5.4.44 source included Zend/bench.php and Zend/micro_bench.php files.
• 3 runs are made and average taken for time it takes to run the benchmarks.
• This test is single threaded in nature so doesn't scale with multiple cpu cores/threads and the underlying performance is based partly on cpu processor performance and clock speed. Linode London > Linode Singapore > Vultr London > Dediserve London > DigitalOcean Singapore > Dediserve Singapore > DigitalOcean London

PHP configure options

 

DigitalOcean vs Linode vs Vultr vs Dediserve - UnixBench Benchmarks

August 24, 2015 update: Seems i mixed up the numbers for Linode Singapore E5-2680v3 KVM column was switched with Linode London E5-2680v2 KVM so bare that in mind

As per below notes, you can only validly compare UnixBench system index scores within the same branch of Linux kernels 2.6.x or 3.x or 4.x and also same virtualisation type XEN vs KVM. Comparing 2.6.x vs 3.x or 4.x kernels or Xen vs KVM won't be directly comparable due to differences in subtest numbers that the index score is derived from and other virtualisation specific factors. However, KVM is proven to be faster than Xen even for the same server. So Dediserve is at a disadvantage here.

Vultr VPS kernels

Code:

uname -r
2.6.32-504.30.3.el6.x86_64

DigitalOcean VPS kernel

Code:

uname -r
2.6.32-504.12.2.el6.x86_64

Linode KVM Kernel

Code:

uname -r
4.1.0-x86_64-linode59

Dediserve Xen Kernel

Code:

uname -r
2.6.32-504.12.2.el6.x86_64

To compare, you can probably only rely on looking at the Dhrystone and Whetstone numbers to gauge cpu processor performance. Dediserve is rolling out KVM later on, just not available at time of testing so Xen VPS was used. However, for pure performance, Linode is the winner.

Notes regarding UnixBench Index numbers

You can read the explanation at byte-unixbench - A Unix benchmark suite - Google Project Hosting which suggests subtests like pipe throughout has no real world bearing on performance so can be disregarded

 

@JarylW and @RoldanLT will be interested in Singapore results and @Matt will be interested in London results

 

I must of got lucky with Xeon E5-2680v3

 

Intel Xeon E5 v3 vs v2 is about 10-20% faster clock for clock, so it roughly translates to 0.2Ghz to 0.35Ghz. So E5-2680v3 at 2.50Ghz is roughly similar to E5-2680v2 at 2.80Ghz unless specific tasks or tests don't take advantage of the architecture improvements in the Haswell-EP based Xeon E5 v3.

Where E5 v3 will shine in theory is for SSL, encryption and decryption performance.

 

Also updated Centmin Mod Benchmark page with this thread's listing Benchmarks - CentminMod.com LEMP Nginx web stack for CentOS

 

You mean for centminmod.com and underlined hyperlinks ? yeah it's part of the default style it removed underlines from hyperlinks. I'll slowly add them back eventually but would have to have separate css for it as other areas need to have no underlines for hyperlinks i.e logo and index box content listings.