Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Hetzner EX 4S Unixbench, dd and htparm results

gist.github.com

Post reply on HN
Print view View original

31–40 of 45 posts

Re: Hetzner EX 4S Unixbench, dd and htparm results

#31

Post by mjb »

Travis wrote: post #27
Diederich wrote: post #26

Earlier quoted context omitted.

If you are lucky, processes would crash. More likely there would be silent bits of data corruption.

Gotcha, so it would just depend on what bits in RAM were corrupted (data versus process instructions)? I am curious about a couple of specific use cases that I can think of where this might affect me (and, most likely, would be common points of failure for others): 1. data in the MySQL tables (stored in memory) is corrupted. Would mysql crash? Indicate table corruption, and I could just reload it from disk? Write cor…

The answer to most of your questions is 'maybe', unfortunately. Reasoning about the things that could happen when memory errors occur is very difficult, because they occur outside the mental model of computation that most programmers (and systems administrators) use.

Let's use MySQL as an example. A bit flip in the memory which holds the code may cause it to crash. A bit flip in the 'metadata' could cause the table to become corrupted, potentially recoverably. A bit flip in the data itself could turn 'Travis' into 'Trbvis', which might go undetected depending on where it happened and which storage engine you are using.

The use of memory for OS page caching (less so in databases, which often use O_DIRECT and more so in other programs) means that arbitrary corruption could happen to pieces of disk data your program didn't even touch, if you touch data near them.

Re: Hetzner EX 4S Unixbench, dd and htparm results

#32

Post by jules »

Travis wrote: post #27
Diederich wrote: post #26

Earlier quoted context omitted.

If you are lucky, processes would crash. More likely there would be silent bits of data corruption.

Gotcha, so it would just depend on what bits in RAM were corrupted (data versus process instructions)? I am curious about a couple of specific use cases that I can think of where this might affect me (and, most likely, would be common points of failure for others): 1. data in the MySQL tables (stored in memory) is corrupted. Would mysql crash? Indicate table corruption, and I could just reload it from disk? Write cor…

If it was a C int then it would just change to another value. If it was a Python int then two things can happen: either the bit flip was in the value which causes the value to change, OR the bit flip was in the tag bits which causes Python to interpret the data as something else than an int. The latter would most likely cause your program to crash.

With MySQL any of those things you can happen. If you're lucky then only the cache is corrupted and then you can just reload from disk. If you're unlucky then the data got corrupted on its way to disk and the wrong data will be written to disk. If you are astronomically unlucky then the in memory machine code of MySQL got changed in such a way that it starts overwriting your entire disk with garbage. You should probably be more afraid of meteorites though. And of bugs in either your own or others' code.

ECC RAM reduces the probability of such a bit flip happening. That doesn't mean that they are eliminated entirely. So you have to do these two things in any case:

1. Bit flips can cause processes to misbehave/crash. So you want to have a way to detect and restart misbehaving/crashed processes.

2. Even with ECC RAM you want to do your own error correction for critical data (say a bank transaction log).

Here is an interesting paper that discusses the prevalence of DRAM errors and the effectiveness of ECC RAM:

DRAM Errors in the Wild: A Large-Scale Field Study -- http://www.cs.toronto.edu/~bianca/papers/sigmetrics09.pdf

It would be interesting if somebody did an experiment where they artificially flipped bits of various software's memory to see what happens. I'd expect that in many cases it doesn't do any harm at all.

Re: Hetzner EX 4S Unixbench, dd and htparm results

#33

Post by nirvana »

vladd wrote: post #7

The price for Hetzner's EX 4S is 49.58 EUR for non-EU customers and 59 EUR per month VAT-included for most EU customers (EU corporations registered for VAT purposes need to apply the VAT specific to their own country). An Amazon m2.4xlarge instance (High-Memory Quadruple Extra Large Reserved Instance) with Heavy Utilization on a 3-year commitment costs $9'660 for the 3y term and 3 * 365 * 24 * $0.454 = $11'931 in usa…

The only problem with your comparison is that you forgot to include the 149 Euro setup fee for the Hetzner server. I don't think this makes a material difference however over three years.

Some would say this isn't fair because you're not including elastic scaling which AWS supports, however you can't elastically scale with a 3 year commitment. Further its hard to estimate how much scaling you'll need because that's a highly variable figure.

But you can say that you will need to keep some number of servers up and running 24/7 for your service.

I think it would be a more perfect comparison to compare AWS's on-demand price to Hetzner with the built in higher performance of the 4S considered as a hot standby for scaling. (I don't know how long it takes Hetzner to provision extras, though I hear its relatively fast.) Since servers are cheap (at Hetzner, at least) it seems reasonable to have 1 or 2 extra spun up if you really have spiky traffic.

Since I'm planning to get a cluster of Hetzner boxes to support a Riak distributed cluster, and each Hetzner box is over provisioned for my needs (if I get a 4S I think they will be way over-provisioned, leaving a lot of headroom for this "elastic scaling" issue)... I think I'll compare the price of the same number of each boxes. (The AWS box has less performance, but I can elastically scale if I need to, and I'll just give the cost of those short term scale-ups to Amazon for free in this example.) The results may not be a perfect example for whatever others are considering, but it is a reasonable comparison for my purposes. (my service is a big data service, involving a lot of map reduce, requiring a cluster for operational simplicity and reliability as well as the ability to scale to millions of customers. I don't know if we'll get millions of customers, but one of our closest competitors had millions of customers within a few weeks.)

Amortizing the 4@ m2.4xlarge = 4* $600/month = $2,400 (using the $600 for the reserved instance costs from vladd's example.)

4@ Hetzner 4S = $149 Euro setup fee, amortized over 3 years = 4.14 Euro/month Add the 49.58 Euros in hosting = 53.71 euros x 4 servers = 214.87 euros or $282.50 a month.

So, for my cluster, AWS is 8.5 times as expensive.

I think that $282 per month is a lot easier for a scrappy startup to handle than $2,400 as well. This brings "big data" within the realm of undefended mortals.

Many people use AWS for all the extra services that AWS offers. The value of these services is, of course, subjective, as some startups will need them and others won't. But they all are based on a custom Amazon API which means if you build your architecture across many servers at Amazon you've got a fair amount of lock-in.

For me, I'll take the 8.5X reduction in costs, over lock in anyway. If Hetzner starts performing poorly, I can migrate without too much difficulty, and without having to re-architect my service.

Re: Hetzner EX 4S Unixbench, dd and htparm results

#34

Post by inovica »

krobertson wrote: post #15

Why is this company getting so much attention as of late? I've never heard of them until recently, and considering how crowded the whole VPS/dedicated server market is, what makes them stand out?

I've used them for a few years now. I think they used to be very much German-language focused, but over the past couple of years they have become more and more international. They give good support and the servers I have with them have been rock solid

Re: Hetzner EX 4S Unixbench, dd and htparm results

#35

Post by nirvana »

prateekdayal wrote: post #29
dasrecht wrote: post #16

Earlier quoted context omitted.

Of course Elastic Scaling is not possible. Because Hosting a server in a datacenter and having cloud instances are two different stories. If you have need for flexibility you can simply run a loadbalancer having your base hardware hosted for a good price at hetzner (or another datacenter you like) and then spin up some AWS Instances behind the loadbalancer for dealing with spikes. Or as others have written you can bu…

The problem is that you need to have all the three boxes on the same rack wired up to a gigabit switch. Hetzner can do that for you but you can't add or remove servers easily. I have found that OVH has a virtual rack that can accomplish this. After years with Hetzner I think it is time to move on

Do you have data for the difference in latency &/or bandwidth between servers on the same rack at Hetzner vs. on different racks?

I was assuming that having the servers in different racks would be an advantage, mainly because the reason I have multiple servers is for data redundancy (I'm using Riak and data is replicated across the cluster.) However, in Riak the servers do talk to each other a fair bit.

The latency and bandwidth between a given server and a web browser out in the world shouldn't vary much from rack to rack, I'm assuming.

Re: Hetzner EX 4S Unixbench, dd and htparm results

#36

Post by julian37 »

fleitz wrote: post #19
julian37 wrote: post #13

Earlier quoted context omitted.

That's hardly a fair comparison. Last time I checked Hetzner doesn't offer auto scaling, "elastic" load balancing, or any of the countless other features that EC2 offers. Actually, forget auto-scaling: what if you want to bring up another bunch of instances manually to deal with a spike in demand? The EX 4S looks like a very compelling alternative for many use cases but saying that its 9x cheaper is comparing apples…

Actually it's a really fair comparison, for the price of the AWS service you can buy 9 hetzner instances. So you don't need to spin anything up, just buy it. A long term contract is the perfect comparison because it's a similar service, you know you'll be using the capacity so you pay up front to have it 'reserved' and in exchange receive reduced prices. Normally with EC2 you'd buy 1 instance and then spin up 8 more…

You and your sibling posters are all focusing on my point about scaling. Fair enough, maybe the fact that it's so much cheaper means that point is moot.

That still doesn't mean that it's a fair comparison. EC2 has plenty of other features out-of-the-box like the ability to manage security groups, VPNs, elastic IPs, easy access to other AWS offerings like SQS and S3, the ability to treat images and the machines they're running on as independent, multiple locations, etc.etc.

Yes you can have all these things outside of the cloud, and yes you might not need any or all of them. (EDIT: and yes using their services means you're locked-in to some degree.) Still, I maintain my original point which is that you can't really say that Hetzner is X times cheaper based purely on the hardware specs, disregarding the fact that EC2 is much more than just a bunch of virtual machines.

Re: Hetzner EX 4S Unixbench, dd and htparm results

#37

Post by mike_esspe »

nirvana wrote: post #35
prateekdayal wrote: post #29

Earlier quoted context omitted.

The problem is that you need to have all the three boxes on the same rack wired up to a gigabit switch. Hetzner can do that for you but you can't add or remove servers easily. I have found that OVH has a virtual rack that can accomplish this. After years with Hetzner I think it is time to move on

Do you have data for the difference in latency &/or bandwidth between servers on the same rack at Hetzner vs. on different racks? I was assuming that having the servers in different racks would be an advantage, mainly because the reason I have multiple servers is for data redundancy (I'm using Riak and data is replicated across the cluster.) However, in Riak the servers do talk to each other a fair bit. The latency a…

Sometimes i had a 5-10% ping loss between servers in the same Hetzner datacenter, so not being on the same rack can be a problem. Especially if you are using nginx -> backend. Then every request is a TCP connect to the backend server, and when SYN packet is lost, you will have a 3 seconds timeout (at least on FreeBSD TCP stack).

Re: Hetzner EX 4S Unixbench, dd and htparm results

#39

Post by espennilsen »

hjalle wrote: post #30
stephenson wrote: post #28

Earlier quoted context omitted.

As said other places, they have ECC servers to: http://www.hetzner.de/en/hosting/produkte_rootserver/ex6

I never meant to criticise, I saw that they have ECC-servers as well, I just wondered what the implications would be if running a database on a non ecc server, as I've never really understood how much damage it could cause.

Short answer: ECC will save you from having corrupt data written to your database if one of the DIMMs are corrupt.

Re: Hetzner EX 4S Unixbench, dd and htparm results

#40

Post by pnathan »

jules wrote: post #32
Travis wrote: post #27

Earlier quoted context omitted.

Gotcha, so it would just depend on what bits in RAM were corrupted (data versus process instructions)? I am curious about a couple of specific use cases that I can think of where this might affect me (and, most likely, would be common points of failure for others): 1. data in the MySQL tables (stored in memory) is corrupted. Would mysql crash? Indicate table corruption, and I could just reload it from disk? Write cor…

If it was a C int then it would just change to another value. If it was a Python int then two things can happen: either the bit flip was in the value which causes the value to change, OR the bit flip was in the tag bits which causes Python to interpret the data as something else than an int. The latter would most likely cause your program to crash. With MySQL any of those things you can happen. If you're lucky then o…

I suggest looking into studies of radiation effects upon computer systems. They do a lot of bit-flipping. I was privy to results from a confidential study once, and as one might expect, enough bit flips cause big problems (the study went into more details than that, of course).
Post reply on HN