New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
BuyVM - Allegation of Trouble, Lies, Slabs, Hosts Servers in Basement
This discussion has been closed.
Comments
100% accurate. Remember, I do a lot of research, much of the research I do a lot of others use (vswap being broken for 6+ months for instance and an easy fix). Your current deployment is based off my research and findings.
I share lots of my research as well, usually the stuff that I've long since found no use for. I did load tests on both VMWARE, & XEN back in the day.
Besides, you're talking about the guy that went out of his way to stir trouble between people. Do you really believe every last thing he said? I know back then (when that convo went on) you didn't.
The only XEN boxes we had in production in either company were the original 3 we had at Frantech and then the single we had testing + addons01 on.
Francisco
tl;dr
nested-virt + buyvm = true --always
hypervisor during 2010 = ESX
hypervisor now = KVM
did they do slabbing ( eg cram 4+ over a single node) = yes on ESX (until 2010 or later )
do they slab nodes now = no as per @Fransisco ( 1 node = 1kvm)
Desperate Jarland's struggle to accept the facts makes me laugh.
If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck.
Would you like to play child games or would you like to help me connect the logic? I'm reading what he's giving me and it isn't supporting the words he's saying afterward. Are you reading what I am or have you already made up your mind and you just feel like insulting anyone who doesn't subscribe to your personal prejudice?
Incorrect good sir.
On ESX we did a single 8GB node per physical node.
Remember, BuyVM was created as a 'market tester', we never had any idea the brand would grow very fast. Frantech still had a fairly good amount of VMWARE based customers so we couldn't go cramming that much usage into a dual quad w/ 24GB ram.
Francisco
@Francisco
I'm sorry but the log I posted earlier in this thread completely contradicts what you've stated here.
Please quote the exact text that contradicts. I beg you, I'm not seeing it, I want to see it haha.
"On ESX we did a single 8GB node per physical node."
The image I provided you with earlier contradicts this. I'm quite sure I have logs of him stating exactly how much he put on ESX, and it certainly was not 1 8GB slab per physical box. Remember, these logs are from 2010 (the beginning of their deployment).
ONLY if you read "on vmware you can have 24G on the node" as "on vmware we have 24G on the node"
Quite frankly, it doesn't say that. That screenshot never says they do, it says "you can." This is a huge difference. Do you have another screenshot that perhaps you're thinking of and just thought it was that one when you linked it?
Do not need one, you're reading the wrong screenshot.
http://imgur.com/lDmXn7F
>
Oh ok, gotcha. Looks like a contradiction. Care to explain @Francisco?
As I see it @Francisco lied in sometime "probably" around 2010 when he said that he only had 3-4GB on a vmware and tonight he said 8GB. So let that be known to anyone who is buying BuyVM, that @Francisco lied probably sometime in 2010 about how much ram was allocated to a vmware.
Fran's reason for running the systems this way is a legitimate one. Adaptec cards don't generally play nice with older kernels, which is one of the main reasons why the majority of providers use LSI cards right now.
BuyVM seems to have many happy customers who are getting at least adequate performance, regardless of how things are setup... so other than not disclosing every detail of his node setups, there's really nothing to discuss here. Rehashing old stuff from years ago accomplishes nothing.
Where?
Dedup consolidates...that's the purpose. XEN didn't have dedup, only ballooning. Why would we go and roll out a platform that we lose one of the strongest features of?
Used & allocated are 2 different things, especially in dedup. The nodes were given 8GB but with merge down from from users in there as well as regular FT customers usage was more around 2 - 3GB on the node side.
Remember, you're talking about a 24GB node here in VMWARE. VMWARE I don't think even allocates swap to be used.
Francisco
Are you implying that he's grown up man who doesn't lie anymore and we should blindly trust him everything he says now?
I stand corrected then, my apologies.
I'm just a referee calling the points. Interpretation is up to the reader.
Wow, kinda old news that it's slabbed on again/off again. I didn't like these stupid threads when they were about CC et al, I still don't like it here. If it doesn't cause performance issues who cares. Get over the righteousness and stop trolling.
Only on the Internet can you fabricate a controversy, provide no arguments of substance, and then claim victory.
If @Francisco has been lying to everyone here then we need to get it out in the open. BuyVM has had a lot of sales from this site and customers deserve to know the truth, whether that be what is already known or what is yet to be known.
You must be new here.
You know, if BuyVM was a ColoCrossing company, this thread would be set to sink...
Interesting discussion. @Francisco, how many physical servers did you have in Batavia, and what kind of connectivity?
16.
It was bonded Verizon business FIOS lines.
We rebalanced things as time went on so there was no saturation issues.
Francisco
That's not really true. If you cut a server into smaller "slabs" then you get better scheduler efficiency because there are less PIDs in each scheduler domain. A similar effect can be accomplished using cgroups, but things like SolusVM do not support that at all. Hince why slabbing is interesting to some which want to overcommit the nodes. It can "soften the blow" so to speak by providing another level of isolation.
I'm still checking out my logs, but here's what I have for you so far. Take it as you please and see if you're still corrected.
http://imgur.com/2iCfJvk
http://imgur.com/A7b3ywV
http://imgur.com/fDHnvln
I believe I have some from early 2011 where he stated the actual number per vmware box, but we'll see when I get that far.
You must be very rarely around.
You know, we see CC threads on the frontpage all the time but, if BuyVM was a ColoCrossing company no one would care much about evidences and @jarland would as example vigorously attack BuyVM right this moment ;-)
Just saying.
On a tangent, Someone go run that script on a HostGator dedicated machine. Fairly sure all of their dedis are KVM layered.
To be fair I remember seeing Fran disclose their use of it for the reasons listed in a dew other dated threads.
Thanks Francisco. Where in Batavia, and which plan did you use? http://www.verizon.com/smallbusiness/products/business-FiOS-Internet/packages/
1 & 3 sound like theoretical conversation, 2 is interesting but lacks context. He said he helped you with your nodes, you didn't deny that, so is he talking about buyvm nodes or yours?
Actually, overcommiting IO is the bigger sin. Righteousness over the 'lying' is extremely subjective and touchy given that you personally wouldn't even know the truth if it came and bit you. All you ever have here is heresay, usually taken out of context, out of time, and with minimal relevant background information, and usually provided by people with an agenda. Filling in the blanks with your own interpretation doesn't make it any truer - and time and time again, everyone has been proven to plug their ears to anything that doesn't already jive with what they already believe is the truth.
All we know as customers, and what we actually can measure ourselves is how the node/pseudo-node behaves. Does it OOM? Does the IO wait spike?
I say again, much ado about whether @Francisco decided to slab a few nodes smacks of nit-picking petty jealously. Especially given your line of "BuyVM has had a lot of sales from this site". You yourself work for the competition despite not having the gumption to stick it in your sig. If you're so uptight about transparency and truth, do it now.
If slabbing (as done here) really is an issue, let the chips fall where they may and let loose the hundreds of my BuyVM VPS constantly OOMs posts proceed.
P.S. I make the same rants against all the frivolous anti-CC crap so you can rest assured I'm not backing anybody because they're favorites.