Reserving Disk Space you Purchased from the Oversold VPS providers

[Francisco]

okay that's a catch all phrase. I wonder what @Francisco 's stand on this matter

edit: how do you tag a user here?

"Don't be a dick".

We have lots of users that use all of their RAM and diskspace. We overbuild our nodes these days just to give extra room. Infact we're chatting right now about bumping disk & bandwidth in the summer after a trip to Vegas to pump up the nodes some.

CPU is a fair share basis. If you're running all 4 cores on your VM? you're going to get capped. Monbot handles that all and gives us amazing reports like this:

[09:00:26] ny-node04 says:

Process ID 215916 (java) from CTID 22131 using 139.64% CPU. Triggered action cpulimit (80%)

We plan to merge those reports into stallion 2 so end users have access to them.

Francisco

 

[jarland]

For me, abuse should be something objective.


boss @Francisco what is your stand?

It is though. It is a very bad idea to have policies that are so direct and targeted that they have to be adjusted every time someone finds a new way to abuse something. A good policy outlines an idea or concept and tries to confine it with limitations as best as possible while leaving room for application to new forms of abuse when discovered. We are in a market that changes often enough for this to be a big deal.

But keep in mind that my servers aren't meant to be toys for people to try to mess up. They're for hosting real content and services. When someone comes along and just wants to play games with my business plan just for the sake of doing it and seeing if they can negatively impact myself and my other clients, how would I not be forced to view this as abusive intent?

 

[john]

Jarland put it very eloquently. A key advantage to virtualization is over provisioning (overselling) resources to more effectively use said resources and reduce cost.  When hosts start doing a 10:1 ratio using SSD as swap, that's when it gets ugly. Even if your host does not oversell on RAM or disk, your bandwidth commitment is oversold by A LOT. Overselling done responsibly is not bad. 

 

[jcaleb]

"Don't be a dick".

i cant remember. but is that phrase really in the agreement i checked "agree" when i signed up with you guys? i just find it cool.

 

[Francisco]

Jarland put it very eloquently. A key advantage to virtualization is over provisioning (overselling) resources to more effectively use said resources and reduce cost.  When hosts start doing a 10:1 ratio using SSD as swap, that's when it gets ugly. Even if your host does not oversell on RAM or disk, your bandwidth commitment is oversold by A LOT. Overselling done responsibly is not bad.

They already have though.

There has already been one person that has stated CVPS for instance had a pair of SSD's in RAID0 for swap just to try to keep up with memory usage.

Francisco

 

[jarland]

Don't mind me and my legal beagle nose

Please append that policy to include disk... CPU, disk and RAM resources.  Guess I created the appending of disk

Bit perplexed though how using services constitutes abuse!?!  All of those items are quarantined and boxed in by default.  No user can exceed their disk allocation in any way.  RAM is limited to account settings.  CPU is limited too.

 " to a degree that effects the performance of other customers on your node."

One could say that any use constitutes effects on the performance.

All theory and nitpicking.  Friendly chat.

I'll file that in my list for the next update. I think I'm adding bittorrent soon (will be dated, excluding signups before that date) so that'll be a good time.

As for how it would be considered abuse, I think intent is the focus here. I can't prove intent, so I'm not seeking people out and asking them why they're using their resources. However, if they told me that they are forcefully reserving 100% of their allotment purely because they believe I oversell and they want to make sure they have what they bought, with no intent to use it for any actual use, I would call that borderline malicious intent. It isn't painful, but when they're attempting to make it painful for me then I think I'm looking a client who might be happier elsewhere and I'm happy to give them a vzdump full of empty block files

 

[jcaleb]

I wonder how many hosts uses e5 with 128gb ram

 

[john]

There has already been one person that has stated CVPS for instance had a pair of SSD's in RAID0 for swap just to try to keep up with memory usage.

 

I was referring to CVPS subtly. How well does SSD as swap to substitute for real RAM even work, I'd imagine things would become very slow..

 

[Francisco]

I wonder how many hosts uses e5 with 128gb ram

Probably not many.

Even the E5's we've built we only did 64GB in them. We didn't increase our load outs just because we have more resources to abuse. The limits we keep now are predictable. We know on average how many inodes are going to be used, RAM usage, actual bandwidth usage, etc.

Francisco

 

[drmike]

There has already been one person that has stated CVPS for instance had a pair of SSD's in RAID0 for swap just to try to keep up with memory usage.

 

Broken record mode:  Multiple people were presented the idea of using SSDs for expanding RAM by Colocrossing (when inquiring about renting servers).

It's a bad and evil idea.  Someone right now is working to improve the concept.  

 

[Francisco]

I was referring to CVPS subtly. How well does SSD as swap to substitute for real RAM even work, I'd imagine things would become very slow..

Sure.

On most linux boxes you'll get 2 - 3GB/sec in memory bandwidth (I usually test by writing a file in /dev/shm). 2 SSD's may give you something in the 1GB/sec mark. You're going to burn them out hard though if you start smacking on them hard.

Francisco

 

[Francisco]

i cant remember. but is that phrase really in the agreement i checked "agree" when i signed up with you guys? i just find it cool.

Nope but we've had a TL;DR section for a long long time:

Resource Abuse

- Affecting other clients will affect your tenacy.

- You are responsible for `poor code`.

- `Accidents` happen once. And only once.

Francisco

 

[jarland]

I wonder how many hosts uses e5 with 128gb ram

Good ol' Taylor

(no overselling allowed here)

CPU model :  Intel® Xeon® CPU E5-2620 0 @ 2.00GHz

Number of cores : 24

CPU frequency :  2000.169 MHz

Total amount of ram : 129033 MB

 

[Marc M.]

I wonder how many hosts uses e5 with 128gb ram

@jcaleb we do and they are amazing machines, very robust

 

[jcaleb]

Good ol' Taylor (no overselling allowed here)

sure you could, thats a lot of ram

 

[acd]

Just curious, how many providers would object if I ran, say UML in an OVZ? That would pretty much behave exactly like an eat-resources daemon, except doing potentially useful things.

 

[jarland]

Just curious, how many providers would object if I ran, say UML in an OVZ? That would pretty much behave exactly like an eat-resources daemon, except doing potentially useful things.

I mean as long as it didn't hammer I/O and CPU all the time for no reason I'd say sounds fun!

 

[kaniini]

I wonder how many hosts uses e5 with 128gb ram

When I was working for Enzu (after they bought SIP's virtualization business), BudgetVM was running e5's with 192gb RAM.

I'll let that sink in a bit.

 

[Marc M.]

I'll let that sink in a bit.

@kaniini we run e5 2620 and e5 2609 nodes with 128GB RAM with SSD caching (Cache Cade Pro 2.0) for both Xen and KVM, and they are great. From 128 to 192 it's just one step IMHO small nodes aren't worth the money anymore, especially for Xen.

[edit]

Microrant:

It just kills me when I see providers adverise "RAID 10 Protected Storage" - as RAID 10 is meant for high IOPS with minimal redundancy. Speed it's the only purpose why it exists, because otherwise it's the most inneficient way to group multiple drives into an array as you loose 50% of your cappacity no matter what and kind of play lottery because "what if the wrong two drives fail?".

Hmm... I have to run some tests with SSD cached RAID 50. See how it performs under high load. Anyway, rant over :lol:

 

[jcaleb]

There has already been one person that has stated CVPS for instance had a pair of SSD's in RAID0 for swap

that much oversold?