VPS Speed Test - TekTonic - VPS Hosting

techjoe · #1 12-13-2009, 07:00 AM

I have noticed some alarming slow speeds from my VPS at the Texas data center. Not sure if it's being over sold or what but it is enough to cause concern!
I reported that several times on the live chat and there was some other user that was slowing the server down but it's still slower than it should be.

Now I have a server speed test:
-----------------------------------------
<html>
<body>
<?
$Multiplier = 400;
$time_start = microtime(true);
for ($i=0; $i<$Multiplier; $i++) {
for ($j=0; $j<$Multiplier; $j++) {
for ($k=0; $k<$Multiplier; $k++) {
}
}
}
$time_end = microtime(true);
$time = $time_end - $time_start;
echo 'Test took ' . $time . ' seconds';
?>
</body>
</html>
-------------------------------------------------------

On my tektonic VPS at the Texas data center I always get an 8 second response.

But on some other servers that I have at other locations I get 5 seconds and 4 seconds.

Is this an accurate test?

matta · #2 12-13-2009, 07:56 PM

What host node are you on? We keep a pretty good eye on server performance so I'm surprised it's so bad for you.

That test provided measures CPU performance only, in my experiences it is disk I/O that causes customers the most problems (lag). It is entirely likely you're seeing better CPU performance at other providers -- we cap CPU usage for plans to help provide better consistent performance (less burst, but the CPU you pay for is more likely to be there when you want to use it). That is how we get the Max CPU listed for all the VPS plans. If you provide the host node I can certainly provide you the graphs for your node though so you can see for yourself if host CPU is over-utilized or it is the cap in effect.

techjoe · #3 12-14-2009, 04:54 PM

I don't know what host node I'm on. Where would that info be?

It looks like the speed issue was handled because there was some other user that was using up too many exim processes.
The site is faster now.

But that caused me to look into the speed tests.
I installed Zen Cart for a client and it became obvious that it was slow.
Zen Cart has several hundred MySql queries to display a page.
Speed would not be much of an issue if it was a plain HTML page.
Speed with MySql would involve both the CPU and Disk I/O I assume.

matta · #4 12-14-2009, 07:27 PM

Well, the issue with mail servers is that they use a lot of sync() calls which lock the filesystem, it's tough to deal with but we keep implementing technology to bypass it (latest is a sysctl to disable fsync() calls, but that is potentially a problem if a node loses power somehow).

Anyhow, for future info our performance graphs are at http://graphs.tektonic.net.

techjoe · #5 12-19-2009, 07:24 PM

Well, it may sound logical to you to cap cpu speed for your various packages.
But to me it means that sites hosted on that VPS are slower than shared hosting sites at big shared hosting places like dreamhost or lunarbreeze.

Viewed that way it bothers me to offer hosting that would have inferior CPU speeds than shared hosting.
So for that reason I am considering dumping that VPS and transferring to a faster CPU VPS.

matta · #6 12-19-2009, 07:45 PM

The CPU cap really shouldn't decrease performance too much for normal workloads. It's not like you're only getting a couple hundred mhz, the cap is the "Max CPU" number listed on our plans page, so even a $15/mo VPS can use up to 2.4Ghz. We used to not cap CPU for a long time, about 4 years, and I can say without a doubt that doing so definitely provides better service for the vast majority of customers. If you think about it, it doesn't really make sense to give a lot of CPU (~10Ghz) to a $15/mo account.

Anyhow, if you give me your IP I can tell you your host node, that way you can correlate with the graphs and will be able to tell if it is the cap being hit or insufficient resources available on the host node.

Also, in regards to shared hosting, I was never convinced VPS is necessarily supposed to be faster than shared. No doubt a single instance sharing resources with others on a single node is going to be a lot slower than running on a shared cluster with load balancers and multiple web/db/mail servers powering them. The advantage of VPS is the control given over shared hosting. For just hosting websites shared may be better, but if you need specific modules, versions of software, configuration options, etc your options are VPS or dedicated. Of course that's just in relation to web hosting, the flexibility of VPS is really the only option once you go towards application hosting.

techjoe · #7 12-20-2009, 01:26 AM

Ok, the IP number for that was sent by private message.

matta · #8 12-20-2009, 04:31 PM

You're on hsvz21.dal -- http://graphs.tektonic.net/Colo4Dall...dal/index.html

It's definitely not idle, but it's avg of 243% CPU idle for the week (400% max for 4 CPU's). It's quad 2.4Ghz so that leaves roughly 5.2Ghz avail on avg. Since you're on a Linux-720 you are capped at 200% which on this node is 4.8Ghz, so you should be able to get that for your tests.

I also ran the UnixBench v4.1-wht benchmark tool on the node and results were really good:

Code:

==============================================================
BYTE UNIX Benchmarks (Version 4.1-wht)
System -- Linux hsvz21.dal.tektonic.net 2.6.18-028stab066.10 #1 SMP Fri Dec 4 15:49:04 MSK 2009 x86_64 x86_64 x86_64 GNU/Linux
/dev/sda2             10080520   6868688   2699764  72% /

Start Benchmark Run: Sun Dec 20 11:58:19 EST 2009
 11:58:19 up 1 day, 16:30,  1 user,  load average: 3.91, 3.87, 4.82

End Benchmark Run: Sun Dec 20 12:10:19 EST 2009
 12:10:19 up 1 day, 16:42,  1 user,  load average: 26.41, 15.62, 10.62


                     INDEX VALUES
TEST                                        BASELINE     RESULT      INDEX

Dhrystone 2 using register variables        376783.7 38044622.1     1009.7
Double-Precision Whetstone                      83.1     2145.2      258.1
Execl Throughput                               188.3     6411.5      340.5
File Copy 1024 bufsize 2000 maxblocks         2672.0   172114.0      644.1
File Copy 256 bufsize 500 maxblocks           1077.0    53293.0      494.8
File Read 4096 bufsize 8000 maxblocks        15382.0  1256905.0      817.1
Pipe Throughput                             111814.6  4596333.9      411.1
Pipe-based Context Switching                 15448.6  1105170.7      715.4
Process Creation                               569.3    18284.0      321.2
Shell Scripts (8 concurrent)                    44.8     1773.9      396.0
System Call Overhead                        114433.5  6094499.0      532.6
                                                                 =========
     FINAL SCORE                                                     497.3

So I'm kind of stumped why your cart is loading so slow.

One thing I would check is your MySQL tuning. I notice you have default my.cnf so using the medium sample config should help. ie. cp /usr/share/mysql/my-medium.cnf /etc/my.cnf and then restart mysqld.

Also, what is the load time of your cart site? I usually use http://www.iwebtool.com/speed_test to measure this.

techjoe · #9 12-20-2009, 07:40 PM

I had tried a large mysql cache.
This had mixed results like
Size Load Time Average Speed per KB
15.08 KB 13.91 seconds 0.92 seconds
15.08 KB 0.67 seconds 0.04 seconds
15.08 KB 0.43 seconds 0.03 seconds
where the first hit took the longest and then it speeded up.

I changed to the medium my.cnf without a cache and things are more even now.
15.08 KB 0.67 seconds 0.04 seconds
15.08 KB 0.84 seconds 0.06 seconds
15.08 KB 0.4 seconds 0.03 seconds

That must have been one of the problems along with the exim processes.

Thank you for taking the time to troubleshoot this.