Featured Vultr Benchmarks 2GB KVM VPS location benchmarks for Vultr 6-way compare

Just launched 6x 2GB Vultr KVM VPSes in 6 different location to run my centminmodbench.sh comparison benchmarks like I did here. Since I am developing my own custom GeoDNS based SPDY SSL CDN using Centmin Mod web stack as the core, I thought I'd see how these specific Vultr locations would do

Referral links:

If you like this comparison review, please consider using my referral links to sign up or put the word out and share this benchmark review with your friends etc :

• Vultr - offers free $5 credit for new sign ups

Vultr KVM VPS Specs

• 2GB Memory
• 2 CPU cores
• 40GB SSD
• 3TB Bandwidth
• CentOS 6.5 64 bit

Vultr KVM VPS Locations Tested

The following 6x Vultr KVM VPS locations were tested:

• Los Angeles
• Dallas
• New Jersey
• London
• Tokyo
• Australia

Initial Virgin centminmodbench.sh benchmarks

Initial tests are done on a fresh CentOS 6.5 64bit OS without Centmin Mod LEMP web stack. This is to test the systems without any interference from Centmin Mod LEMP web stack's installer and tweaks to TCP etc.

I use SecureCRT SSH client which allows sending a single command to all launched SSH windows, so basically I ran centminmodbench.sh simultaneously on all 6x Vultr KVM VPSes, so benchmarks reflect at the same period of time for bandwidth speed and ping results.





Tabulated centminmodbench.sh virgin run times. Fastest to completion was New Jersey at 708 seconds followed by Los Angeles at 727 seconds and slowest was Sydney Vultr KVM VPS location at 1,125 seconds.

Vultr Disk ioping tests

Default ioping performance is pretty much the same across all but one Vultr KVM VPS location - London was nearly half the IOPs compared to other locations and cached IOPs was nearly a third.

Vultr Disk dd tests

Same trend, all Vultr KVM VPS locations pretty close except slower at Vultr London.

Vultr Disk fio tests

---

Again same trend with disk fio performance for random reads and writes being closer on all Vultr KVM VPS locations except London which is nearly half to a third the speed in terms of MB/s and IOPs.

Vultr Bandwidth Speed tests

Bandwidth download speed tests are geographically relative, so closer the download site to the Vultr KVM VPS location, the faster it is usually. Unfortunately, it seem in the case of Vultr Sydney VPS, while it was the fastest, it wasn't that much faster than some West coast USA locations.

Los Angeles, Dallas and New Jersey look pretty good across all bandwidth download locations.

Vultr Ping tests

Ping times are also geographically relative. Los Angeles does look like a good all rounder though.

Vultr OpenSSL Benchmarks

• OpenSSL system benchmarks test against system default OpenSSL 1.0.1e for number of ciphers and configurations would be good indicators for https SSL performance given Google now includes https in their ranking signal search engine algorithms.
• Relevant tests include rsa 2048bit, rsa 4096bit and 256bit ecdsa key algorithms. You can read why the newer ECC SSL certificates using 256 bit ecdsa key algorithms are better performing than the more commonly used rsa 2048 bit keys here. As well as test aes128/256 cbc non cpu hardware assisted performance and AES-NI cpu hardware assisted performance via the evp aes128/256 numbers.
• Again looks like Vultr London VPS performance is lagging behind the rest at less than 1/4 of the performance compared to the rest of the other Vultr locations.
• Best rsa 2048/4096 bit performance goes to Vultr Tokyo followed by Vultr Dallas, New Jersey, Sydney and then Los Angeles with London at last place.
• There's a change in leader board when it comes to 256 bit ECDSA signatures with New Jersey falling behind at almost 1/2 the performance for signatures and verifications. London again came last though.
• hardware assisted AES-NI aes128 and aes256 performance has Vultr Tokyo in the lead again and Vultr London last.

 

Centmin Mod .08 beta01 Install Times

Same with Centmin Mod .08 beta01 installation, using SecureCRT to send the command to launch the CLI installer.sh method of installation to all 6x Vultr KVM VPSes simultaneously.

Centmin Mod .08 beta01 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.

The CLI installer.sh shows both the YUM portion of time as well as total Centmin Mod install times. So roughly, total install time minus YUM install time gives you an idea of source compilation and other source download times.



I'm going title this next screenshot - waiting for London Basically, the London Vultr KVM VPSes must be pretty loaded and congested as Centmin Mod .08 beta install times were one the slowest I have ever experienced !. Check the tabulated results for a better breakdown of the times.



Fastest overall time was Vultr KVM VPS in Los Angeles at 827 seconds or 13.78 minutes followed by Dallas. If you take into account only the non-YUM portion of the install times, then Dallas was fastest followed by Los Angeles. The slowest was London at 2,777 seconds or over 46 minutes !



The new Centmin Mod .08 beta01 default Nginx installed index.html page



The Centmin Mod .08 beta01 menu where all the magic happens

Code:

--------------------------------------------------------
Centmin Mod 1.2.3-eva2000.08 - http://centminmod.com
--------------------------------------------------------
                   Centmin Mod Menu              
--------------------------------------------------------
1).  Centmin Install
2).  Add Nginx vhost domain
3).  NSD setup domain name DNS
4).  Nginx Upgrade / Downgrade
5).  PHP Upgrade / Downgrade
6).  XCache Re-install
7).  APC Cache Re-install
8).  XCache Install
9).  APC Cache Install
10). Memcached Server Re-install
11). MariaDB 5.2, 5.5, 10 Upgrade Sub-Menu
12). Zend OpCache Install/Re-install
13). Install ioping.sh vbtechsupport.com/1239/
14). SELinux disable
15). Install/Re-install ImageMagick PHP Extension
16). Change SSHD Port Number
17). Multi-thread compression: pigz,pbzip2,lbzip2,p7zip etc
18). Suhosin PHP Extension install
19). Install FFMPEG and FFMPEG PHP Extension
20). NSD Re-install
21). Update - Nginx + PHP-FPM + Siege
22). Exit
--------------------------------------------------------
Enter option [ 1 - 22 ]
--------------------------------------------------------

 

centminmodbench.sh + Centmin Mod .08 beta01 install benchmarks

Re-running centminmodbench.sh but this time after Centmin Mod .08 beta01 LEMP web stack is installed. With Centmin Mod .08 beta01 LEMP web stack installed, additional tests are detected and triggered in centminmodbench.sh as well as TCP settings tuned.

These additional tests include:

• Centmin Mod's Nginx static OpenSSL 1.0.1i benchmarks
• Axel parallel download speed tests (to compare with the single threaded wget bandwidth speed tests)
• mysqlslap MySQL benchmarks
• PHP Zend/bench.php and Zend/micro_bench.php benchmarks.

Post-Centmin Mod centminmodbench.sh Times

Additional tests do add additional time to complete centminmodbench.sh. Vultr Dallas was fastest to completion followed by Vultr Los Angeles while Vultr London, New Jersey and Sydney lagging behind. But Tokyo was slowest at 3,815 seconds.

Centmin Mod's Nginx static OpenSSL 1.0.1i benchmarks

Axel Parallel Download Speed Test

mysqlslap MySQL benchmarks

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 acorss 64 clients.

Actual test command used is:

PHP Zend Benchmarks

• PHP benchmarks test the Centmin Mod .08 beta01 installed PHP 5.4.33 (php-fpm) binary (which defaults include enabling APC Cache 3.1.13 with igbinary support) and tests against the PHP 5.4.33 source included Zend/bench.php and Zend/micro_bench.php files.
• You can see the PHP 5.4.33 default compiled extensions here.
• 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.

Disk ioping tests

Disk DD tests

Disk fio test

Bandwidth speed tests

Ping tests

 

Bandwidth Numbers Compared

Comparing pre and post Centmin Mod installations, bandwidth numbers there's some gains but quite a few slower losses - will need to tweak my TCP settings more to get it right.

Bold next to each pair, marks which was faster pre or post Centmin Mod.



Let's check out centminmodbench.sh's gathered mtr stats for just Vultr New Jersey VPS to see if we can see if there's differences that can explain the differences in bandwidth speeds.

First up is mtr side by side for Cachefly, Linode Atlanta and Linoda Dallas download locations.

Cachefly for 2nd post Centmin Mod tests had much higher ping times averaging 47.2ms with last ping at 83.6ms compared to pre Centmin Mod test average of 22.3ms and last ping at 14.5ms.



Vultr Atlanta also showed differences with higher average ping times of 33.4ms vers 22.8ms and last ping 40.8ms vs 22.6ms. A lot of hops went through slightly different location it seems



You can see rest of locations side by side for Vultr New Jersey VPS.

 

Feedback and questions welcome

 

Yeah Vultr London performance was poorer, probably the host node was busier ?

 

who knows.. will re-try testing again with newly launched Vultr VPS

 

better late than never to reply to this.. benchmarks are all done using my centminmodbench.sh script at bench.centminmod.com - I have included a pre-made Google Spreadsheet you can use to fill in the benchmark scripts outputted results.