Post by trog @ 04:20pm 23/03/10 | 129 Comments
The AusGamers website has actually been running off a couple of load balanced Mammoth VPS instances for the last couple of weeks, and we've been really happy with the performance - we're still serving tens of thousands of people a day (and now we have the advantages of redundant infrastructure!).
If you're interested in a virtual server of your very own, check out Mammoth VPS at http://www.mammothvps.com.au/. As always, we're keen for any comments, thoughts or suggestions! For those interested, we have an special introductory offer for AusGamers members - 40% off for the first 2 months.
Check the comments for the top secret code, but note this offer is only valid until 31st August, 2010!
Latest Comments
129 Comments
Show
Follow Us
Advertisement
Log In
Posted 05:06pm 23/3/10
AGN40
Valid until 31st August, 2010!
last edited by trog at 13:19:53 02/Aug/10
Posted 05:15pm 23/3/10
+1 for virtual environments, it's going mainstream in a big way recently. On the desktop it makes software dev so much easier - workstations don't getting trashed with various testing garbage.
Posted 05:17pm 23/3/10
Posted 05:44pm 23/3/10
Posted 05:49pm 23/3/10
The idea being:
- we demonstrate confidence in the product by using it for our own services
- we are exposed to most of the same service problems that customers might experience which will help us be less ignorant to customer issues
- gives us first-hand, ongoing experience on the performance and features of the service
Posted 05:56pm 23/3/10
I assume its vSphere?
Posted 06:15pm 23/3/10
The hardware so far is dual quad core xeon 5530's with 48gb of ram. Memory is allocated and dedicated to your guest, it's not oversold
Posted 06:32pm 23/3/10
Posted 06:58pm 23/3/10
Posted 06:58pm 23/3/10
Posted 07:06pm 23/3/10
bah: what blacklist?
Posted 07:06pm 23/3/10
Posted 07:12pm 23/3/10
but I guess it depends on the eventual implementation
Posted 07:13pm 23/3/10
Purchase a local VPS, a US or EU VPS and setup a vpn between them, route your traffic over the VPN. TADA! filter bypassed.
Easier ways than that to do it too.
Posted 07:21pm 23/3/10
Posted 07:23pm 23/3/10
Posted 07:32pm 23/3/10
pARODY 1+
last edited by HerbalLizard at 19:32:56 23/Mar/10
Posted 07:40pm 23/3/10
The bandwidth limits would take care of that easily, anyone that wanted to do mass seeding would be way better off with a slice or similar. Thats what I like about this plan, it doesn't include stupid amounts of data, so less chance of chumps hogging the tube. Going to give this a shot cause slice is great, but a bit too much latency for my liking.
Posted 07:46pm 23/3/10
Posted 07:48pm 23/3/10
Posted 07:48pm 23/3/10
F*** moving it again... :)
Posted 08:00pm 23/3/10
Yeah you can just vpn to a non-au host/vps/colo and gain the same thing. Using a local one will allow you to setup some routing trickery so that your local .au traffic can stay in .au with decent latency or go over the VPN for your non-au traffic. Plus it can mask your actual source address to be the server for that added layer of privacy. :P
Bouncing is the only way to travel online ;)
Posted 08:56pm 23/3/10
- By default do you get an imaged Linux install like Ubuntu or something to get you up and running?
- If you need Windows do you license it yourself or is licensing through Mammoth with a floating pool of Windows licenses for example?
- What happens if you hit your data allowance, is it charged at a certain rate after that or just cut off?
- Is there a service to dump a whole bunch of data on your VM like AWS Import/Export? Amazon have extra charges for it obviously.
Khel:
If you're just after basic svn hosting then there's a few cheap/free sites out there that do just that to if you wanted to save the hassle of setting up a repository and maintaining it yourself: beanstalkapp.com or unfuddle.com, or if you're doing open source you probably couldn't beat code.google.com. As you say, Ventrilo would probably be a good usage for VPS i'd reckon, then again Ventrilo is really really cheap elsewhere as well. Might make sense if you have a bunch of diverse services you want to host.
Posted 09:09pm 23/3/10
Posted 09:16pm 23/3/10
- yep you get a default minimal image - the choices atm are debian 5, centos 5.4, ubuntu 9.10, fedora 12 and windows server 2008 standard
- the windows licensing is included in your monthly subscription and yep we provide the license (the key is already installed into your default image)
- data over allowance is charged at [some rate I'm not sure of atm sorry]
- nope sorry
Posted 09:24pm 23/3/10
http://110.232.115.153/vpsdemo/
In terms of other features we dont have let us know if theres something you think we need we're all ears.
Posted 09:27pm 23/3/10
Posted 09:32pm 23/3/10
As per the sign-up page:
Posted 09:37pm 23/3/10
That's actually pretty good pricing. I remember being stung by telescum at $0.35/mb.
Posted 09:42pm 23/3/10
Posted 10:14pm 23/3/10
Posted 10:19pm 23/3/10
You can think of each plan as having a daily charge (eg the $30 plan is $1 a day), so if you spend 10 days on that plan and then spend the remaining 20 days on the $60 plan the total cost to you for that month will be $50 (10/30 x $30 + 20/30 x $60)
Posted 10:21pm 23/3/10
Posted 10:23pm 23/3/10
Posted 10:28pm 23/3/10
main reasons would be that it's free, easy to integrate with an automated site, and currently pretty much the best performing hypervisor for paravirtualised guests. plus the hardware support is quite broad - with xen it's more about the hardware your chosen linux distro supports than anything. vmware and xenserver installers for example, don't even support ich fakeraid - which I find quite frustrating. it means you have to buy quite specific hardware to use them and/or chew up precious pci slots, and even more specific hardware to be eligible for support
Posted 10:30pm 23/3/10
Ah I see, fixed billing dates. I was mixing up the upfront new plan charge with plan changes.
Cheers, I was going to ask for some screenshots but this did the job 1000x better.
Good work guys!
Posted 10:39pm 23/3/10
actually on the subject of traffic - that reminds me that the default australian repo for ubuntu (maybe debian as well?) is actually hosted overseas and on quite a poor-performing connection for australian clients last we checked. several of our staff already using the system commented on it thinking it was our network that was slow. after they changed the repo to bigpond or something else actually local, it was fine. we're actually about to set up some local mirrors for the distros we provide and when we do that, we'll update our default images to use those instead, and make the info available to existing customers as well. this should hopefully be done either next week or the week after pending the time it takes to trickle the mirror data into the network without congesting it.
Posted 10:47pm 23/3/10
Cool, we havent really done anything with that on the site yet, was just something I knocked together, might do a shortened version at some point and link it to the current screenshot
Posted 02:17am 24/3/10
The site says that backup options are planned in the future. Are snapshots on the drawing board too? I've lost count of the number of times I've taken a snapshot in VMWare just before making a large change/installation, borked the system up completely and been saved by being able to roll back cleanly to a stable image.
I can think of a few ways to implement this:
1. Initial idea: allow a maximum of one snapshot per plan at no charge. Effectively doubles storage requirements, probably not feasible.
2. Better idea: allow purchase of snapshot "slots" for frequent, long-term snapshot use.
3. Another idea: allow temporary (e.g. 24 or 48-hour expiry) snapshots at no charge. Can only use this feature X times a month, non-cummulative. Great for occasional use and minimises long-term storage requirements.
To be honest I think offering both #2 and #3 would be pretty cool.
But hey these are just some late-night ideas .... my finger is twitching to sign up and test this system out.
Posted 04:32am 24/3/10
What's your SAN ?
Posted 04:52am 24/3/10
Any local data count? Any upstream services as the reason I choose the current vps provider was that they had a squid farm upstream to address caching issues but didn't address overselling the slices and local i/o performance issues.
I think its awesome that Mammoth are heading down this path and also awesome that they are offering up decent discount to ausgames members, wonder what they have planed in the future.
last edited by HerbalLizard at 04:52:21 24/Mar/10
Posted 05:18am 24/3/10
We've recently VM'd ( vsphere enterprise) a bunch of servers at work ... replaced over 20 servers with 3. With the advanced VM options they self load balance, fail over, etc between the ESX servers.
Posted 09:50am 24/3/10
Unfortunately domestic and resonably priced don't usually fit in the same sentence.
Posted 10:13am 24/3/10
Oh also where are the seedbox plans?
last edited by Mass at 10:13:49 24/Mar/10
Posted 10:19am 24/3/10
Our data priceing is pretty good for the Australian market, but its definitely no comparison vs the US - if you need a lot of bandwidth, its much cheaper to host overseas.
The RAM limits are hard set; Xen doesnt support bursting. Excess data is charged at 50c a gigabyte; the HDD size is also a hard limit - though you can change plan and get a bigger HDD in a few minutes.
Posted 10:23am 24/3/10
Posted 11:25am 24/3/10
haha ISWYDT
Posted 11:32am 24/3/10
Posted 11:38am 24/3/10
Thecus? :p
Posted 11:40am 24/3/10
Posted 12:08pm 24/3/10
What are some cool things I can do with this thing? I guess there is always irssi.
Posted 12:20pm 24/3/10
http://blog.mammothservers.com/20100215/hosting-a-wordpress-blog-with-a-mammoth-vps
(note: blog is not sposed to be online at the moment but that was the start of a sample thing I wrote for it. I think I have all the steps in there but let me know if you get stuck)
Posted 02:21pm 24/3/10
Edit: I'm going to add some more questions and repeat the other one;
1) Is data counted in/out/both
2) Will you be introducing load balancers
3) Will you be introducing firewalls (specifically supporting IPSEC VPNs)
4) Will you be introducing storage services outside of a VPS (ie: connected via iSCSI)
.. that's all for now :)
Edit Edit:
5) Any SLA?
Posted 03:15pm 24/3/10
2. We don't have it automated yet, but we do support the use of our load balancer by clients
3. This is getting a bit technical for me to answer as I dont really know anything about IPSEC - I presume you are referring to a separate firewall in front of your VPS?
4. All our storage is provided via iSCSI, but that probably doesnt answer the question?
5. We dont have any specific service guarantee. As Jim mentioned though, we are running AusGamers on the exact same product.
Posted 03:33pm 24/3/10
6th question - do the VPS have a private IP?
Posted 04:10pm 24/3/10
Posted 04:13pm 24/3/10
Jim is away for a few days, when he's back I'll find out about the firewall question.
The guests dont have a private IP by default (or an automated setup yet), but we are happy to setup this up by request. We currently use this between the load balancer and AusGamers web servers
Posted 05:18pm 24/3/10
Posted 05:29pm 24/3/10
Posted 05:31pm 24/3/10
Posted 05:38pm 24/3/10
Posted 07:23pm 24/3/10
Posted 02:06am 25/3/10
Nice and simple process, setting up apache/php etc now.
All done running centos with virtualmin / webmin to configure services and websites (have about 5 relativley low usage.
I'm paying a us company $20 od a month for far more than I use might aswell support Aussies and get some benfit from reduced latency and still ample resources Nd having much more flexibility in what I can do.
last edited by 3x0dus at 02:06:25 25/Mar/10
Posted 03:40am 25/3/10
Posted 02:23pm 25/3/10
Posted 02:34pm 25/3/10
Posted 02:53pm 25/3/10
Not yet. Everyone internally keeps asking for this feature too :)
Posted 03:11pm 25/3/10
Posted 03:13pm 25/3/10
Posted 03:18pm 25/3/10
Theres no way I'd move to a virtualised service that didn't provide at least a previous-day backup of the service image. Most of your competitors offer this for free guys, and have similar pricing (at least for the enterprise packages).
You're gonna need to do it if you want business in that sector :)
Posted 03:23pm 25/3/10
Posted 03:26pm 25/3/10
Its a component of a backup solution.
If the s*** hits the fan in a bad way, you are back in minutes. How long (ie, $) will it take to recover a 50G system from offisite backups?
Posted 03:27pm 25/3/10
Posted 03:48pm 25/3/10
Point is, I don't consider it a show stopper as you should have your own backup system in place. Wether this is pulling your images back to your office (srsly?) or setting up reundancy in the VMs and having your own data backed up to another VM or cloud based storage. Currently I'm using a mixture of GoGrid & RS and have load balanced frontend servers, clustered DBs, hourly transaction logs shipping to local disk, copied to a another VPS at both vendors and then also shipped back to our office here with daily backups. Content is replicated locally to each VPS, to the DB servers and also (at GoGrid) to their cloud storage.
I'll take that over getting an image of my VM done daily.
Posted 04:22pm 25/3/10
Posted 04:27pm 25/3/10
So you mustn't have been talking to me then I guess.
Posted 04:40pm 25/3/10
Posted 04:42pm 25/3/10
Posted 04:49pm 25/3/10
Not every business has a need to send hourly transaction backups into the cloud, or will be inclined to pay for that level of management. Why would someone with these requirements be housing their server on a budget VPS?
Posted 05:31pm 25/3/10
As mentioned we're looking at providing backup options; I suggested during development that people might want to be able to download their own VM image entirely as a backup (I know I want to!) and that we offer a 'backup to EC2'-esque service as an offsite backup offer. We're still taking about what backup features to offer though so any feedback you guys have about what sort of backup services you'd require for your exact business needs would be hugely appreciated.
Posted 05:33pm 25/3/10
Posted 09:13pm 25/3/10
Posted 09:41pm 25/3/10
1) Daily snapshots. Not as a full backup solution, but if something screws up in the image I have a fast initial restore point.
2) Free bandwidth to 1 IP. Not to use as a seedbox or for any illegal reasons (still be defeated by bw limits anyway) but so I could implement my own backup and transfer solution myself without having to worry about bw limits and extra charges.
3) DNS, although this is more a nice to have.
Posted 09:57pm 25/3/10
What exactly would be wrong with that? A decent blade chassis will have 4 psu's, at least 2 network switches, up to 6, probably 2 fibre channel switches and will hold 14-16 blades, which could be dual quad core machines.
A blade chassis like that would have over 100 cpu's cores and could probably run several hundred vm's pretty easily.
Also, a decent san will have redundant everything.
What do you think runs most datacentres?
Posted 11:17pm 25/3/10
Posted 11:42pm 25/3/10
1) Redundancy.. there isn't any?! I need at least a daily backup to consider putting my sites on there.
2) Can I buy multiple C class IP's from you?
Posted 11:45pm 25/3/10
Posted 06:07am 26/3/10
I'm a bit confused about Mass's comment, and wonder what he expects the envrionment to be but a bunch of servers and a san? We've built some pretty hit hard environments and they have all been, in their base form, exactly that. So Mass, any chance you can expand on what your talking about there.
Posted 08:30am 26/3/10
Even the Internet has blinking lights.
Posted 08:45am 26/3/10
----------------
I moved across my US based CPANEL/WHM Reseller account to Mammoth VPS in less than a few hours, using GPL Virtualmin/Webmin.
http://www.webmin.com/
Steps below!.
1. Sign up for a VPS!
- Choose Centos 5.0
2. SSH into VPS
3. wget http://software.virtualmin.com/gpl/scripts/install.sh
4. sudo chmod +x install.sh
5. sudo install.sh
wait, after a few minutes it will be installed and you can access the virtual/webmin server @ https://yourserver:10000/
now by default it only uses PHP 5.16 which some may not like
if so.
6. sudo rpm --import http://software.virtualmin.com/lib/RPM-GPG-KEY-virtualmin
7. sudo rpm -ivh http://software.virtualmin.com/bleed/centos/5/i386/virtualmin-bleed-release-1.0-1.rhel.noarch.rpm (for x86) or
sudo rpm -ivh http://software.virtualmin.com/bleed/centos/5/x86_64/virtualmin-bleed-release-1.0-1.rhel.noarch.rpm (for 64)
8. sudo yum update
9. on the https://yourserver:10000/ goto system information click on status click on restart apache.
Apache will restart and PHP 5.2.11 will be in use.
Its recommended for compatibility to stick with what Virtualmin provides, but obviously you can mix and match at will.
Now the easiest thing is to import your websites from a Cpanel host.
Login to old cpanel host(make sure you know the password else reset the password before doing the backup), goto backups Do full backup
this will get saved to /home/server/backup.xxx
chmod this for read access, and move it to /home/server/public_html
on your Virtualmin server goto Webmin in the top left corner.
Click on "Others" on the left, click on "Upload and Download"
enter in the address to the backup file http://oldsite/backup.tar.gz etc
save it to somewhere on the hdd.
if you wish you can also download the file abit more *safely* in ssh via
wget -c http://oldsite/backup.tar.gz
Once its download to import this to your new server with pretty much everything in tack (subdomains/mailbox/ftp accounts/etc etc)
1. back to https://yourserver:10000/ (if not already click on virtualmin) on the top left.
2. click on "Add Servers" click on "Migrate Servers"
3.it should be set to local file already, click the [...] browse to the place where you saved the backup.tar.gz select it.
4. Password for administrator put the dot in and type the password.
Click Migrate.
Wait a few, and bam its all done, check the log data make sure everything came across, on occasions the MYSQL file wont import properly but that simple matter to import via Virtualmin interface also. You may also need to fix some folder permissions.
So in a couple of hours your fully setup and moved across to free alternative to cpanel with alot of customisable options, aswell as some good backup system built in.
last edited by 3x0dus at 08:45:15 26/Mar/10
Posted 10:10am 26/3/10
1) Are all the eggs in the one DC or do you have another site?
2) Do you have redundant hardware or at least an SLA with a Vendor and if so what is the response time? (this was kinda answered that you have multiple servers)
3) As Nats said - if you were/are to provide backups will they be offsite?
4) I know you've said you're not providing an SLA but in the event of major disaster what is your process?
I'm still keen to sign up.....so take my Amex and I'll take a Silver Plus to kick off and then (after evaluating it properly) another Silver Plus and a Gold Plus.
You guys should also check out Microsoft Hyper Server 2008 R2 (the free version). It's an excellent piece of kit. I've got 2 boxes running it with 12 guests, 3 Oracle servers, Exchange 2010, 2 Ubuntu boxes, 2 Webservers etc. Clustering and live migration work a treat and they are piss easy to manage.
Posted 09:20pm 28/3/10
- there are currently a couple of single points of failure in our system, in the interests of maintaining a reasonable price point for the initial product and target customer base. I'll provide more info on this tomorrow, time pending
- as already mentioned, the topic of backups is quite varied - we're hoping to be able to cater to as many people as possible here, and I can't see any technical reasons why we can't
- ipsec firewall: as it happens the infrastructure in place can technically provide it, but it didn't even occur to me that someone would want it. I've got a couple of questions about that myself
thanks for all the feedback and questions
Posted 01:21am 29/3/10
I have three virtual switches,
VMSwitch1 is attached to VMNIC1 is attached to internal-only VMs ...
VMSwitch2 is attached to VMNIC2 is attached to a a single centos machine running Quagga ( routing protocols )
VMSwitch3 is not attached to any NIC and is attached to the quagga machine and all of the DMZ hosts.
Quagga announces the DMZ network running on VMSW3 via OSPF. It uses IPSEC firewall to restrict access to only certain ports on certain IP/VMhosts.
no machines on the DMZ have virtual nics on both DMZ & private VMSwitches.
With everything set up the same on my VSphere cluster I can have the physical host die and they'll all automatically be brought up on one of the other VSphere servers with full state etc so nobody loses sessions or anything.
so you could extend this to a single customer and give them two virtual machines, one with access to a physical NIC, the other connected only via a virtual NIC to the first. route a /30 between them and you're good to go.
Posted 08:39am 29/3/10
HUH, wrong answer, hyperv doesnt come clost to vmware or xen in anything.
Posted 08:47am 29/3/10
08:46:24 up 2 days, 16:32, 2 users, load average: 0.00, 0.00, 0.00
Well done! Will definitely look into this for all of my sites over once dns/backups are sorted.
Posted 02:18pm 29/3/10
going to go through these questions and give answers one by one as I go:
currently a linux server with 2 x quad core xeon 5520's, 12gb ram, using a battery-backed adaptec 54225 for the raid and i/o, configured in raid6 with a hotspare as well, so that rebuilds can begin immediately upon failure instead of sometime after we invoke our onsite support. this host has 10 x 1gbit nics (2 onboard, 2 x quad port pci - all intel 82576 chipsets) wired into a 2-switch stack. storage is presented to each vps host via iscsi, currently using the chelsio software target. we use linux round-robin bonding to achieve greater than gbit for a single tcp connection. the host also has an ip kvm and redundant power supplies, each plugged into a separate dedicated ups, which in turn plug into separate power feeds provided by the datacentre. however, the host itself and the adaptec controller are single points of failure at this time. depending on how the service picks up, we will address these single points of failure by using another server and probably drbd, as well as a second controller in each and 2-port adaptors on the drive enclosures.
here's a screenie showing a hdtune file benchmark on this storage when it was running windows during initial testing
Posted 02:02pm 29/3/10
If it's more bang for buck I'm sure he'd jump ship.
Posted 02:28pm 29/3/10
I think we'll get this addressed fairly soon - there's no technical reason we can't provide this, in fact our use of lvm should actually make snapshots fairly trivial at least, however some of the other areas of backup would probably require separate additional storage to negate the impact of hardware failure, so might be bit more of a hurdle to some extent, in terms of keeping initial cost down of a brand new service which hasn't proven it can earn it's keep yet =]
Posted 02:33pm 29/3/10
Posted 03:06pm 29/3/10
specifically, the vm hosts are a medium-density type server (2U quad) - so there's a 2U chassis which contains 4 individual servers which are similar to blades, but not as tightly-packed, and only the power is shared via a backplane which runs off two power supplies each fed into a separate circuit as per the storage server I mentioned a couple of posts above. they have their own nics (6 x 1gbit - 2 onboard and 1 quad port pci, intel 82576 chipsets again). so they offer some density, but don't rely quite so much on a shared backplane/enclosure as blades. they each have local raid1 hotswappable drives for their own o/s and as mentioned we provide the storage for vm's via iscsi. 1 onboard nic + 1 of the 4-port pci nics makes up an active/backup team wired into a 2-switch stack for internet connectivity and 1 onboard nic + 3 of the 4-port pci nics makes up the storage team, again using linux round robin bonding mode to stripe single connections across multiple nics. each host has it's own ip kvm
yep everything is in the one data centre. I don't think this will change in the near future either, unless service uptake or demand suggests that the financial outlay would warrant it. I'd love to see it happen, don't get me wrong :) it just needs to make sense from a $$$ point of view.
the hardware isn't truly redundant no. several components of the solution are, but as mentioned above there are a couple of single points of failure. our vendor SLA is 4-hour same day on servers and their components.
offsite backups: to be determined
disaster process: naturally this would depend on which component/s failed, and any number of scenarios could occur, each with varying impacts and responses. however the focus will always be to minimise the customer impact, as obvious as that sounds. so the process might vary from anything between migrating vm's to another server to invoking vendor support to swap out faulty components
Posted 03:26pm 29/3/10
Posted 03:57pm 29/3/10
Posted 04:07pm 29/3/10
The amazon cloud computing is used for data proccessing, you would probably get an API or something to connect to the cloud to proccess the data.
Whereas the mammoth thing is a Hosted Virtual Private server on the internet.
Posted 04:26pm 29/3/10
Posted 09:15am 30/3/10
"Xen VPS Hosting - www.mammothvps.com.au - Linux or Windows hosting in Sydney Try us now with Instant Setup!" gmail advertising :D
I got a Silver+ to cheq it out, 50% off was too good to pass up, so far so good...
Posted 09:29am 30/3/10
As such, you are required to persist all your data "offsite" - Amazon of course recommend using their S3 service, but there are other options. This has a lot of flow on effects about how software that works on Amazon's infrastructure needs to be structured; as mentioned data-intensive processing is a great fit. Hosting your company's web-site, not so much.
Posted 03:24pm 03/4/10
[vps@dom000185 ~]$ uptime
15:13:11 up 7 days, 17:38, 1 user, load average: 0.02, 0.05, 0.03
Posted 06:21pm 13/4/10
Decided we'll support all these options and have started development today. One free snapshot per week with the ability to purchase more on a daily/weekly/monthly basis as desired. Although instead of a "temporary" snapshot, we'll let you replace an existing snapshot any time.
Posted 10:30am 14/4/10
Posted 10:48am 14/4/10
huh? that doesn't make sense. Snapshotting in the context of vmware or SAN, the current blocks are marked read only and then changes are recorded ... so unless you change 100% of your data, there's no way it would take up double the space.
I have several hundred snapshots on my SAN at any give time, some of my more important volumes are snapshotted every 15 minutes. If each snapshot doubled your storage requirements I'd be in a lot of s***.
Posted 10:53am 14/4/10
Posted 10:56am 14/4/10
Posted 11:26am 14/4/10
however, it depends on the context - how long are the snapshots going to be kept? for regularly-replaced ones it's fine but for user-initiated ones you have to either set a timelimit on them or cater to the fact that a substantial portion of the user's blocks might change during the lifetime of the snapshot. so yes, probably not 100%, but a decent amount
the other factor to consider is that the very nature of snapshots means a substantial amount of i/o is added because any time a write occurs on the original, it now has to be checked in the exception table, read, the original data written out to the snapshot and only then can the requested write occur on the original. to that end, we will probably be taking our snapshots and then immediately writing out the image to a new partition and removing the snapshot. so unless we compress them, we will be doubling our storage requirements.
Posted 11:29am 14/4/10
Yes I use VMWare daily and am aware of this. However this isn't in the context of VMWare.
Doing some quick research before I posted that comment, it appeared Xen did not have a proper snapshot feature. You have to dd by yourself. I may be wrong, as I've been reading snippets of a save feature floating around, but can't find anything solid as I'm not overly familiar with Xen.
Posted 11:37am 14/4/10
we're using lvm2 for our vps storage so that's how we would take snapshots
Posted 11:44am 14/4/10
Posted 11:51am 14/4/10
so unless you have some other means of taking snapshots, then yes you'd probably have to just write out the entire (live) image which means you should really shutdown the vps for the entire duration to avoid filesystem inconsistencies
but we do have other means of taking snapshots (lvm) so we don't need to dd the live image - we can just pause the vps, send it a signal to flush write buffers, take a snapshot and then unpause the vps and then work with the snapshot if we need to do anything else like write it away somewhere else. the snapshot process only takes a second or two so the vps isn't usually paused long enough to interfere with stateful network connectivity
Posted 01:11pm 14/4/10
In LVM2's case, the different is pretty unacceptable - up to 10x slower in some tests, for that reason we'll be taking "full" backups. Giving up so much write performance just to converve a bit of disk space isnt a worthwhile tradeoff.
Posted 07:03am 15/4/10
Posted 08:04am 15/4/10
Yes and No ... in my experience LVM snapshots don't suffer from too much performance degradation until the delta gets to be quite large. I think of snapshotting as short term thing, and therefore I wouldn't expect a large delta. This wisdom spreads across to other snapshots ( vmware, zfs, etc ) and to a lesser degrees SAN block level snapshots ( generally handle it much smarter ).
So if you ensure the snapshots have a short life ( a day or two, even a week doesn't normally create a large delta on a basic web server ). What the snapshot does though is allow you to back up off the snapshot without having to worry about the state of the active data ( for databases etc you need to quiesce before the snapshot, but with snapshots taking minimal time this is usually unnoticable ).
So as long as your users understand that snapshot != backup, and you put some smart time limits on the snapshots I would think in 90+% of cases the performance and storage overheads would be minimal. You could even set rolling snapshots to do a daily snapshot and delete the old one. using LVM this would be simple to set up with cron.
Posted 08:18am 15/4/10
Posted 09:19am 15/4/10
Try the test from the page I linked, its perfectly repeatable on our hardware - the instant a snapshot is created, there is a large degradation in write performance. Like 5 to 10 times slower.
If you aren't doing heavy writes (and clearly many people do not - your typical website is pretty lean on writes) you probably would never notice, but from a service provider's perspective we can't assume anything about the workload customers will have. A proper 1:1 copy of course has its own downsides (takes up more disk, takes longer to restore from) but at least has no ongoing impact on the VPS.
Posted 02:18pm 06/5/10
looking for a nice billing system actually for my own clients, anybody got any suggestions? tried whmcs, which im kinda at odds with even tho everyone uses it... the mammoth ordering/billing system looks quite good, like the simplicitiy of the single ordering page.. with a website like that, i actually couldnt resist. you guys using a commercial one rebranded, or a custom one?
Posted 02:28pm 06/5/10
Posted 02:55pm 06/5/10
Posted 03:33pm 06/5/10
Posted 03:25pm 28/5/10
"Each Mammoth VPS customer receives an off-server, on-site backup once a week free of charge. For many customers, this will be a suitable backup solution; however if you absolutely cannot afford to lose data you must ensure you also back up your data to an offsite location.
The number and/or frequency of backups can be increased for an additional fee."
Posted 03:57pm 28/5/10
Posted 05:29pm 08/7/10
(fail2ban blocks after 1 fail for 1 month)