ServerBear Performance Comparison of Rackspace, Digital Ocean, Linode and Vultr
-
-
@travisdh1 said:
@dafyre said:
It is odd to see that DO has the lowest IO rate, but the overall highest score.
IOPS it makes sense. Yes, write is slower, but read is crazy fast. Or do I have the read/write reversed? I'm used to seeing things listing read first
I wrote the header backwards.
-
-
@IRJ said:
@scottalanmiller said:
@IRJ said:
@scottalanmiller Is lower better?
Lower what exactly?
bench
Higher is better. I think that IOPS are a small amount, if any, of that score. It's about computational performance.
-
@scottalanmiller said:
We need some latency numbers from around the world. Anyone want to collect some for us?
Here is the first IP address. A long running ping (hundreds or thousands of pings) would be good, we need the final stats from that:
- 104.236.119.59
- 108.61.151.173
- 172.99.75.133
We have a good idea on bandwidth, IO, CPU and memory. Network latency is pretty huge.
104.236.119.59 = 8 MS Average 1,000 pings
108.61.151.173 = 7 MS Average 1,000 pings
Mangolassi.it (162.242.243.171) = 14 MS average 500 pings -
Whoops, sorry. We just killed the RS node because it is expensive
You can test that RS ping against mangolassi.it instead. Same location, same node type. Sorry.
-
@scottalanmiller said:
Whoops, sorry. We just killed the RS node because it is expensive
You can test that RS ping against mangolassi.it instead. Same location, same node type. Sorry.
With the RS nodes being so expensive... why would you not stand them up on DO or Vultr?
Edit: I mean for production and not tests like this.
-
@scottalanmiller said:
Whoops, sorry. We just killed the RS node because it is expensive
You can test that RS ping against mangolassi.it instead. Same location, same node type. Sorry.
Running now. Will update my above post.
-
@dafyre said:
@scottalanmiller said:
Whoops, sorry. We just killed the RS node because it is expensive
You can test that RS ping against mangolassi.it instead. Same location, same node type. Sorry.
With the RS nodes being so expensive... why would you not stand them up on DO or Vultr?
Edit: I mean for production and not tests like this.
Well DO and Vultr were not well known or well tested at the time that most of the RS nodes were created. And RS still offers a lot of features that those do not, like load balancers. But these days, the advantages to RS are fewer and fewer.
-
@wirestyle22 said:
@scottalanmiller said:
Whoops, sorry. We just killed the RS node because it is expensive
You can test that RS ping against mangolassi.it instead. Same location, same node type. Sorry.
Running now
Thanks
-
Right now, we are favouring a migration to Vultr. But the Linode test is running and is a major contender. Information on that to follow....
-
updated above
-
Ping statistics for 108.61.151.173:
Packets: Sent = 204, Received = 203, Lost = 1 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 58ms, Maximum = 62ms, Average = 58msPing statistics for 104.236.119.59:
Packets: Sent = 231, Received = 229, Lost = 2 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 56ms, Maximum = 66ms, Average = 56msPing statistics for 162.242.243.171:
Packets: Sent = 95, Received = 94, Lost = 1 (1% loss),
Approximate round trip times in milli-seconds:
Minimum = 51ms, Maximum = 56ms, Average = 51msAbout the same from west Texas.
-
OMG WE HAVE A WINNER!!!!
Linode just took everyone out back and took their lunch money!! They have load balancers too!! (a la Rackspace and Amazone.) Look at that IO capacity!!! And that UNIX Bench! Their single thread was by far the fastest too!
-
@scottalanmiller said:
OMG WE HAVE A WINNER!!!!
Linode just took everyone out back and took their lunch money!! They have load balancers too!! (a la Rackspace and Amazone.) Look at that IO capacity!!! And that UNIX Bench! Their single thread was by far the fastest too!
Wow. That's fantastic.
-
@wirestyle22 said:
Wow. That's fantastic.
I'm so excited. No question that they are by FAR the hardest to use, but who cares. That performance is crazy!!
-
@scottalanmiller said:
@wirestyle22 said:
Wow. That's fantastic.
I'm so excited. No question that they are by FAR the hardest to use, but who cares. That performance is crazy!!
Rewarded complexity is fine by me
-
Throughout the range, Linode comes in either as cheap or cheaper than everyone else, too. It pretty much tracks Vultr until it outscales them. Then it matches or beats DO.
-
Of additional consideration... Vultr and RS cap out pretty small. DO and Linode make massive single nodes, which is important when we are running epic databases, which we are doing. The growth rate on the database is quite healthy.
-
