OVH Community, your new community space.

Recommendations for maximum performance


Myatu
12-12-2013, 21:40
Quote Originally Posted by Kode
Can you use the main server ip for a container? Or would you need a seperate IP?
You would have to forward ports using NAT (to a VM listening on a private IP).

Kode
12-12-2013, 20:36
Can you use the main server ip for a container? Or would you need a seperate IP?

elcct
12-12-2013, 10:25
Quote Originally Posted by Kode
Could I not just do a restore over the public 1Gbps connection?
You can. That's huge advantage over SYS servers.

edit:

I mean if you kept backup on the other server. You could also probably have 1Gbps RPN on just one server and proxy data to another....

Kode
12-12-2013, 08:13
Could I not just do a restore over the public 1Gbps connection?

DigitalDaz
11-12-2013, 23:47
The RPN 1Gbps is quite good and I was surprised at the half decent speed of the SAN. Just bear in mind if you went for the less than a gig option, if you did have a situation where you needed to restore, then anything other than gig is going to be time consuming.

Kode
11-12-2013, 23:26
the 1Gbps option would cost around an extra £22.50 per server per month though, so about £800 a year extra, the other possibility if have the RPN 1Gbps was worth it would be just to get 2 servers, and put the small sites on a VM on one of the servers and get either the 200 or 500 rsync space for the backups, though not sure just upgrading the private nework is worth the extra.

*edit* plus at that point you are getting close to the OVH infrastructure range which have better specs.

Myatu
11-12-2013, 23:16
Having the RPN-FE would also open up a whole slew of additional options, such as their SAN storage - you could reduce/change your storage requirements for the images in that case (though you'll very likely want the 1 Gbps option).

Kode
11-12-2013, 22:50
I'm starting to think if OVH don't come out with any decent sized SSD servers in SYS I might just get 3 x DEDIBOX® LT and set them all up with proxmox as 3 x DEDIBOX® LT + 3 x FO IP + 100 domain plesk license comes to about the same as I'm paying now and I could use one of them for the small sites/backup as well as start up an extra instance of the API server should I need to load balance

Myatu
11-12-2013, 21:57
Quote Originally Posted by Kode
The more I look at proxmox the more interesting I find it.
*rubs hands together* another soul taken *evil laugh*

Quote Originally Posted by Kode
Are there any practical differences between linked clones and full clones apart from linked sharing disk space and full requiring their own?
The linked clones store only the differences from that master template. So if you are not expecting a huge amount of differences, then this could mean quite some disk space savings. The downside is that this is only relevant for linked clones on the same host (or partition) where the VM template lives.

How would I deal with things like memcached? Currently I have memcached running on each server, would I install memcached inside the openvz container or on the host?
It depends really. Having Memcache inside the container means it will move with it, should you migrate that container elsewhere. So there's no messing with with accessibility outside of it (network configuration, ensure its running/installed on the host node, etc.). It would also keep it clearly separated from other VMs, which may be required to ensure its security or namespace collisions. If you want to go fancy, consider Couchbase, which can do sharding.

RikT
11-12-2013, 13:11
you would install memcached inside the container, and on the plus side with vm's you can also create a template from your own setup so if required you can just deploy that template and you are already setup ready to go.

Kode
11-12-2013, 10:08
The more I look at proxmox the more interesting I find it.

At the very least it seems it would allow for simple migrations in the future, add a new node, move the vm to the new node, change the master(if old master is being removed), migration done.

I could also set up specific templates, so if API performance was struggling, get another server, add it to the cluster, deploy the API server template. The nice part about that is I could create and test the templates without worrying about affecting the core system.

Are there any practical differences between linked clones and full clones apart from linked sharing disk space and full requiring their own?

How would I deal with things like memcached? Currently I have memcached running on each server, would I install memcached inside the openvz container or on the host?

Kode
10-12-2013, 22:59
many thanks for all the replys guys.

I will give proxmox serious consideration, my only concern is being a newbie at it if I get stuck I don't know whether I will have enough time to get it sorted for when it needs to go live and also being a newbie getting it setup optimally is probably not going to happen.

My current thoughts are assuming 240GB+ drives don't come out for SYS:

Get 2 x http://www.online.net/en/dedicated-s...dedibox-lt2k14 for the API and main website, possibly set up as VMs (openvz or KVMs?) scrap my SP32, but keep the SP16 as with the news that if I order it for a year it will cost 9 months, which puts it on par with what a server from online.net would cost me but has the added bonus of a 30 domain plesk license, so I can use this just for the small sites and backups and not all my eggs will be in 1 basket.

soyouthink
10-12-2013, 17:20
Quote Originally Posted by elcct
What kind of latency are we talking about? Because i doubt it will be a noticeable difference unless your public interface is really congested. Anyway i doubt anyone caring about latency would use OVH's private network anyway, but rather use a DC that is more flexible and can provide private switches etc.
Summary: http://www.mysql.com/products/cluster/faq.html (Question 12)
Replication and clustering is not same thing as we know if you click the link inside the answer it explain with some details.

Also there are some other problems about building clustered structures over public network, like throughput. But this topic is just for basic suggestions for a specific situation I think. so I dont want go far and make things more complex for Kode.

elcct
10-12-2013, 16:53
Quote Originally Posted by soyouthink
Again, Buiding VPN or IPSEC network is not problem. Problem is latency. if the network works over managed public network this cause latency and sure you can share files etc but you cant do clustering that needs low latency network. like loadbalancing request or mysql clustring that need performance. If you build a private network between SYS servers and get milyon request on mysql cluster you can see what I mean. This is why OVH offers virtual rack+private network service or unmanaged UDP ip services.
What kind of latency are we talking about? Because i doubt it will be a noticeable difference unless your public interface is really congested. Anyway i doubt anyone caring about latency would use OVH's private network anyway, but rather use a DC that is more flexible and can provide private switches etc.

soyouthink
10-12-2013, 16:02
If you say so.
As a summary if the subject is performance I say you cant. And I say virtualization dont give you performance. And I wrote my suggestion.
Good luck.

Myatu
10-12-2013, 16:01
Quote Originally Posted by Kode
Thanks for that guide Myatu, i'll give it a good look when I get home.

How do backups work with vms?
You have various options with Proxmox. I had a GlusterFS partition mounted, which I had declared in Proxmox as a backup partition (you can also use NFS or regular partition/mount), and just told it to create a compressed snapshot of a VM each night. This is the enitre VM, so restoring it is also the whole thing.

Myatu
10-12-2013, 15:57
Quote Originally Posted by soyouthink
"This" not means virtualization. I am talking about horizontal scaling . SYS servers do not have private network service and without it the network between servers run with huge latency and clustering not works as expected. This was what I mean.
I didn't refer to virtualisation in my reply? You can build a private network through various means, and is not limited to hardware only. Hence my response "sure you can". The private network runs the same distance as a server-to-server communication, as that's the switch's job, so the latency is negligible using either method (VPN or not). I've run a Galera cluster between three geographically different locations (Milan, Vienna and Amsterdam), and even this latency is acceptable. Btw, you don't get 1:1 increase in performance with an MySQL cluster due to ACID compliance.

soyouthink
10-12-2013, 15:49
Quote Originally Posted by elcct
You can build private network with tools like http://tinc-vpn.org/ and it works quite well.
Again, Buiding VPN or IPSEC network is not problem. Problem is latency. if the network works over managed public network this cause latency and sure you can share files etc but you cant do clustering that needs low latency network. like loadbalancing request or mysql clustring that need performance. If you build a private network between SYS servers and get milyon request on mysql cluster you can see what I mean. This is why OVH offers virtual rack+private network service or unmanaged UDP ip services.

Kode
10-12-2013, 15:46
Thanks for that guide Myatu, i'll give it a good look when I get home.

How do backups work with vms?

elcct
10-12-2013, 15:39
Quote Originally Posted by soyouthink
"This" not means virtualization. I am talking about horizontal scaling . SYS servers do not have private network service and without it the network between servers run with huge latency and clustering not works as expected. This was what I mean.
You can build private network with tools like http://tinc-vpn.org/ and it works quite well.

soyouthink
10-12-2013, 15:22
Quote Originally Posted by Myatu
Sure you can.
"This" not means virtualization. I am talking about horizontal scaling . SYS servers do not have private network service and without it the network between servers run with huge latency and clustering not works as expected. This was what I mean.

Myatu
10-12-2013, 15:12
Quote Originally Posted by soyouthink
You can not build this structure with SYS servers.
Sure you can.

Myatu
10-12-2013, 15:09
Quote Originally Posted by Kode
I installed proxmox on my kimsufi that's not currently doing anything, but I have no idea what I'm doing, lol.

I installed an openvz container from a template, but I have no idea how to even get it to see the outside world, or how the outside world would get to resources on it.
If you don't have an available failover IP, you could do Step 2 in the Poor man's Proxmox cluster, and read up on Step 7 on how to setup a VM.

Step 8 will give an example on how to receive traffic on a particular port. If your VM needs to be able to talk to the outside world, you need to do two things on the host node (not within a VM):

Code:
# To allow VMs on a private IP to access the Internet (via vmbr0):
/sbin/iptables -t nat -A POSTROUTING -o vmbr0 -j MASQUERADE

# To allow IP forwarding:
echo 1 > /proc/sys/net/ipv4/ip_forward

soyouthink
10-12-2013, 14:26
Virtualization is not for performance and it not give you performance but easy management, isolated structures etc.

For your situation you need a horizontal scaling farm with clustering. You need at least 2 servers to build horizontal scaling. For every server that you add to your farm you get x1 performance that you need for your situation. And you do not need backup servers etc. because you already have 2x what you have.

You can not build this structure with SYS servers. To do this with OVH services, considering your bugget, you need virtual private network that comes with INFRASTRUCTURE RANGE. So min. budget is 82€ x 2 / month.

Another option is online.net DEDIBOX LT servers. They come with Private RPN-FE Network. so min. budget is 35.5€ x2 /month.

Below is the basic diagram. If you interested I can give you more information.


RikT
10-12-2013, 13:40
Quote Originally Posted by Kode
I installed proxmox on my kimsufi that's not currently doing anything, but I have no idea what I'm doing, lol.

I installed an openvz container from a template, but I have no idea how to even get it to see the outside world, or how the outside world would get to resources on it.
essentially all you need to do is add the failover ip in the config when you create the container and that doe sit all for you

Kode
10-12-2013, 13:07
I installed proxmox on my kimsufi that's not currently doing anything, but I have no idea what I'm doing, lol.

I installed an openvz container from a template, but I have no idea how to even get it to see the outside world, or how the outside world would get to resources on it.

Kode
09-12-2013, 23:25
Quote Originally Posted by LinuxGam
Just to double check, I assume you mean someone used their resources to DDOS you, rather than you were hacked and your server was used to DDOS an hence cut off?
Yeah, the site was on a VPS at the time, so it affected other people as well, so null routing the VPS was understandable, but still annoying.

LinuxGam
09-12-2013, 21:44
Quote Originally Posted by Kode
We have been DDoS-ed at least 2 times, we were null routed for both of those (they were with another provider) since being with OVH not sure if we have been DDoS-ed but I remember how annoying getting null routed was.
Just to double check, I assume you mean someone used their resources to DDOS you, rather than you were hacked and your server was used to DDOS an hence cut off?

Kode
09-12-2013, 21:01
We have been DDoS-ed at least 2 times, we were null routed for both of those (they were with another provider) since being with OVH not sure if we have been DDoS-ed but I remember how annoying getting null routed was.

@crowarth thanks I actually started on a new responsive design, but not had the time to finish it http://fanart.tv/wp-content/uploads/...ite-so-far.jpg it's a little less busy and confusing with clear calls to action, but whether it will ever see the light of day is another matter. The status page also needs a bit of work, it was just something I knocked together so I could see the status of the servers at a quick glance (the backup server is offline because I reinstalled it the other day as I planned to use it as the backup / small sites server then after reinstalling it found out the free 30 site plesk license wasn't available on kimsufis)

crowarth
09-12-2013, 20:40
that's a very nicely done site! that status page is pretty cool :3

LinuxGam
09-12-2013, 20:29
Quote Originally Posted by Kode
Annoyingly neither OVH nor online.net have 240GB+ SSDs in my price range (online.net don't seem to have higher than 120GB in any price range) and Hetzner (which has 240GB SSDs) don't have have an Anti-DDoS network and null route the server when there is any DDoS, otherwise i'd have already made my decision.
Is it likely that you will be attacked? Has it happened in the past? Unless I had been attacked before or my server's were syncing my Pacemaker, then it's something I wouldn't worry about until it happens and for now treat user experience as way more important. If you have reason to suspect it may happen or it has happened then I totally understand your worry...

rickyday
09-12-2013, 20:05
With regards to disk/network performance there are many many technologies built into some Hypervisors to massively improve performance, for instance RDMA enabled Nics, SR-IOV, SMB Direct, and dare I even suggest Cluster technologies, which is all probably overkill for what you are looking for tbh. and it will all cost money of course.

As you say chepeast way really is just to have a server with a couple of decent SSDs.

But when the time comes at least you know there are many many options when your site is running off a 16 node cluster

Kode
09-12-2013, 20:04
As I've said, I'm not having any issues at the moment, it all works perfectly, but I'm paying £1530 a year for something with less specs than 2 x SYS XEON-1 which would only cost £1007, or £860 from online.net, OVH have said prices are not going to be reduced for older servers so I would be retarded not to change the servers, so as I'm going to change them anyway I'm just trying to make sure I get the best performance I can, I'd be happy paying the same amount for better specs with SSDs, but there is nothing currently in that range.

The current setup utilises memcached and APC as well.

I think LinuxGam is right that disk IO is going to become an issue before anything else, which is why I was talking about SSDs.

Annoyingly neither OVH nor online.net have 240GB+ SSDs in my price range (online.net don't seem to have higher than 120GB in any price range) and Hetzner (which has 240GB SSDs) don't have have an Anti-DDoS network and null route the server when there is any DDoS, otherwise i'd have already made my decision.

LinuxGam
09-12-2013, 19:37
My guess is, if his main response from the API is looking up graphics and then downloading them, then CPU is possibly the least of his worries. My Experience with running on standard dual raided SATA is that disk IO becomes a problem way before CPU or RAM on even the cheaper servers (that come with Xeon and 16 - 32 RAM).

Upgrading to dual RAID 1/0 (Bad idea though I think) SDD server, will almost certainly cut our your IO delay. Or even better if you can get 2 server on the same 1GB/s network then split the IO load between the two both on SSD's and the performance will go through the roof. I am assuming with heavy file downloads and decent RAM/CPU your bottleneck is disk IO?

If not then as Myatu says... what is it written in? I would fully expect to run out of disk IO before the API does CPU.

rickyday
09-12-2013, 19:30
Don't forget with virtualisation you have the option to live migrate running VMs without any downtime, couple that with DR options like Hyper-V Replica and the world is your oyster. Virtualisation isn't perfect for every scenario buy pretty much most!.

Unfortunately being Windows Server user I cannot help with Nix but I am sure there are plenty that can.

My forte is Microsoft Windows Server 2008/2012/R2 sprinkled with some System Center 2012 R2, would be happy to help there.

Even if you don't feel happy moving everything over to virtualisation just yet, I highly recommend setting up a lab and fiddling with this technology, it definitely is the future and not to at least dabble would be a shame.

Myatu
09-12-2013, 19:27
Virtualisation isn't a solution to everything, of course. It might be suitable here, because you can compartmentalise things as you have now (the API, for example). With a number of them on the same server, you'd put that CPU and RAM to better use (decrease the idle time). If a particular VM would require more resources that isn't available on the current server, you simply migrate the whole thing to a more powerful one (and that would take less time than preparing a new server if it was a bare-metal migration).

With regards to the API itself, have you considered something like Python or NodeJS? You might reduce your CPU and memory requirements that way.

LinuxGam
09-12-2013, 19:25
I personally think the biggest plus out of everything is the ease of complete (os included) backups, so when OVH became a disaster I could move 6 VM's to a new host provider in a less than one day and be 100% back to normal with a few DNS changes. Imagine re-installing Linux, windows and other software on hardware and just restoring data, takes a long long time!

Add that to the fact that you can easily segregate Dev/Test/Live to different DB or Web Instances knowing the non-live will use almost zero resources.. Certainly beats setting up Mysql on different ports on the same server, or worse, the same port and different database names and then you find what works in a view or SP in test, suddenly doesn't when you move it to live because of different DB names :-)

elcct
09-12-2013, 19:21
Quote Originally Posted by Kode
If I had plesk on the server with the SSD it would be harder to setup nginx as it doesnt seem to play that nicely with the apache install (though that was probably a PEBCAK error).

@myatu using VMs is an interesting idea, I've played with virtualbox on my home machine, but never tried to set it up on a linux server.
If you don't change your config too much I'd drop Plesk entirely and configure the server specifically for the project. Once you get how nginx works adding new sites etc. could be quicker than actually doing it in Plesk

For playing with VMs i recommend this toolset http://www.vagrantup.com/ - really handy.

rickyday
09-12-2013, 19:14
Virtualise everything!!

If set up "correctly" virtualised roles/workloads work absolutely fine.

LinuxGam
09-12-2013, 19:13
Quote Originally Posted by Kode

@myatu using VMs is an interesting idea, I've played with virtualbox on my home machine, but never tried to set it up on a linux server.
If Myatu is suggesting it, at a guess I think he would advice Proxmox. I use it mostly for OpenVZ and a Windows KVM and can highly recommend it. If you are running Linux/OpenVZ containers they are particularly lightweight and fast!

Kode
09-12-2013, 19:01
If I had plesk on the server with the SSD it would be harder to setup nginx as it doesnt seem to play that nicely with the apache install (though that was probably a PEBCAK error).

@myatu using VMs is an interesting idea, I've played with virtualbox on my home machine, but never tried to set it up on a linux server.

elcct
09-12-2013, 18:39
Quote Originally Posted by avizeke
person hits website at 36ms response.
images load from the other server would delay the website further as its not from the same machine.
This is not true. It doesn't make a difference if it is not on the same machine (unless that other machine is as slow as phone ), actually if you put images on a separate sub-domain, it will let browser load images concurrently and if you set up correctly browser will not have to send unnecessary cookies etc. improving the speeds further.

Kode
09-12-2013, 18:35
Actually the website serves its own images and the api server does the same, the images folder is rsynced every 3 hours iirc. The API is in PHP. Static resources are generally recommended to be hosted sepertely from the site anyway arent they? And served form a cookieless domain or something.

LinuxGam
09-12-2013, 18:30
Quote Originally Posted by avizeke
if your website consists of images mostly and yet your seperating the two... thats not good in my mind at least.

for example

person hits website at 36ms response.
images load from the other server would delay the website further as its not from the same machine.
I am not sure this would slow it down as if their are lots of heavy files it would be better to split it over more servers and guaranteed bandwidth/IO per server. Once the page HTML is loaded, i.e just text markup (almost instant) then everything downloads Asynchronously, so more disk/CPU IO and bandwidth should improve the overall page download speed assuming both servers are the same distance from the viewer.

Myatu
09-12-2013, 18:10
What language is the API written in? Also, you seem to have gobs of memory available - you should be able to split the server up with VMs, leverage both the CPU and memory better, and have the added benefit of syncing/moving things to a backup server in case it fails.

avizeke
09-12-2013, 17:45
if your website consists of images mostly and yet your seperating the two... thats not good in my mind at least.

for example

person hits website at 36ms response.
images load from the other server would delay the website further as its not from the same machine.

The setup of everything on one server would be better than splitting and segregating.

120Gb SSD and a 2 TB HD server would be ideal.

120 SSD is primary web server with plesk etc and that only.
and then the 2 TB HD for your images , now getting the 2 TB in raid 0 would be best.

and then after this get a backup server

this way you have now limited it to two servers and there are no bottlenecks for waiting for other hosts to respond by sending the images to the website.

Kode
09-12-2013, 15:12
Yeah, 90% of the space is images.

The other thing to remember is SYS is limited to 2 per customer per month or something.

elcct
09-12-2013, 15:04
Quote Originally Posted by Kode
Thanks elcct, interesting idea, although it would probably be quite hard to cache the API requests in any useful way as they are quite dynamic http://fanart.tv/api-docs/music-api/ also having 4 servers might be pushing the budget a bit much, especially when 3 of them would have setup fees

Also I did have nginx setup as a reverse proxy on the main page, but it seemed to mess up a few things on the site (can't remember if it was for everyone or just logged in members, think it might have been something to do with people submitting images) but I will definitely look into it.
I was only saying about images, as i guess these are what use the most of the space. It is very easy to setup with nginx.

As for dynamic stuff like API it needs to be looked at individually as many different strategies can apply.

Kode
09-12-2013, 14:45
The thing is, the current config works quite well as it is, I just know when we previously had it on an SSD server with ovh we could push through a lot more traffic, but having both the API and main site on the same server drew everything to a halt (frequently getting between 10-40 load), splitting them on to their own normal HDD servers brought load right down, but at the cost of potential throughput.

If we have to stick to normal HDDs I would probably be better of just getting the 2 main servers as http://www.online.net/en/dedicated-s...dedibox-lt2k14 which at around £860 for the pair is somewhat cheaper than the £1093.74 that 2x SYS XEON-1's will cost, and significantly cheaper than the £1530 we currently pay.

While that config would only have 16GB RAM as opposed to 32GB I'm not sure it would make that much difference it seems on the server we have with 32GB 26GB is cached, and on the one with 16GB, 13GB is cached.

Kode
09-12-2013, 14:29
Thanks elcct, interesting idea, although it would probably be quite hard to cache the API requests in any useful way as they are quite dynamic http://fanart.tv/api-docs/music-api/ also having 4 servers might be pushing the budget a bit much, especially when 3 of them would have setup fees

Also I did have nginx setup as a reverse proxy on the main page, but it seemed to mess up a few things on the site (can't remember if it was for everyone or just logged in members, think it might have been something to do with people submitting images) but I will definitely look into it.

elcct
09-12-2013, 14:09
You could have another server with normal drives where you store static content and use SSD server as a reverse proxy that will cache most frequently used content. That way you won't need big SSD and still have the speed. Also spare server could be used as an additional backup.

So you could have:
Web - Xeon - 2 2x120GB SSD in Raid 1
Api - Xeon - 2 2x120GB SSD in Raid 1
Storage - Xeon 1 2x2TB in Raid 1

and backup server buy somewhere else, so you don't have all eggs in one basket.

Kode
09-12-2013, 13:36
Thanks avizeke.

The current setup is as follows:


  • Main website (SP32 equivelant to mSP) - apache2 - has plesk control panel and a few other small sites / dev site on it as well - currently around 3TB transfer a month
  • API server - (SP16) - nginx - Currently 8-10TB transfer a month
  • Backup server - (kimsufi) - nginx


My plan is for next year to have 2 servers with 1 provider and the backup server with the other and change the configuration as follows:


  • Main Website (nginx)
  • API server (nginx)
  • Backup server (with plesk / apache2 - small sites plus dev)


I did think about running the 2 x 120GB SSDs in RAID 0, but having 2 disks just increases the likelyhood of 1 of them failing and losing the lot (it will be backed up, but still annoying).

I'm also thinking of having the backup at another location just in case.

If there was a 240GB option on SYS it wouldn't even be a question, I'd snap it up, currently out of OVH/online.net/Hetzner only Hetzner offers 240GB SSDs in my price range, however they don't have an Anti-DDoS network and apparently even smallish DDoS and they will just null route your server, so not really viable.

As for SEO, I do need to do some work on it, but it's hard finding time.

avizeke
09-12-2013, 13:13
very nice website and setup kode. If i was in your position, based on your budget and target on donations etc. Has to be said that the best option for optimised quickloads is SSD. even if you got raid 0 with spinning hd's they still dont come close to load times. BUT this depends on the actual SSD itself. find out its read write times etc.

the SYS options would be best for you buddy, but even those would be overkill for one website i think, having said that i have a MsP just for one website, with 12 mysqls running and alot of scripts and even then its at a low load.


The other option would be to mask IP's and have your website hosting in multiple locations. so that it loads faster to person in different regions. only commercial companies do that though because it becomes complicated and expensive.

as i work in web work and SEO, http://www.alexa.com/siteinfo/fanart.tv i would also look into stratergies of improving your score web value, but as it stands at the moment its good. but being better means you can make a healthier proffit off the website. would also recommend getting .com .us etc and forwarding it to this. that way the domains populate in value also.

Kode
09-12-2013, 11:06
My servers are coming up for renewal, in terms of performance what would be the optimal setup? Everything on a 240GB+ SSD (not something OVH currently does, but who knows if it will come to the game section). Install + database on a 120GB drive + images on a normal HDD (again, not something currently on offer in SYS, but probably more likely than them getting 240GB+ SSDs).

The API currently gets around 56 million hits a month, and in addition to those probably serves 1-50 images for each hit.

The current config isn't really strugging at all http://fanart.tv/status/ but I want to ensure we have room to grow.

The site is currently about 95GB so having everything on a 120GB SSD isn't really practical as it gives very little room to grow.