Market Research - Storage VPS offer tuning.

InceptionHosting

Hi Folks,

I am going to detail a plan below I will not try and justify anything, I am just interested in what you like and dislike and what you would change.

KVM
1 CPU Core
1 GB Ram
1200 GB Disk space (Raid 5 Spinners)
4 TB Bandwidth @ 1 gbit (shared - unmetered inbound)
1 IPv4 address
/64 IPv6 block
Linux/BSD OS supported
Uk/London.

€8.50 /month
€24.00 /3 months
€94.50 /year

seriesn

@AnthonySmith said: 1 CPU Core

May a bit more CPU? Some softwares would benefit from it. Otherwise, prem package sir!

WSS

I was going to say it's a bit high, but that's some decent sized spinning rust. I'd probably consider kicking it down to 768M because some asshole is going to try to run Windows on it.

InceptionHosting

@WSS said:
I was going to say it's a bit high, but that's some decent sized spinning rust. I'd probably consider kicking it down to 768M because some asshole is going to try to run Windows on it.

That just means 2k3 server instead of a secure (ish) one

WSS

Oh god you just gave me a flashback to the very last PC network client I had who was running 2003 on their latest box purchased around 2006; Sage wouldn't work with 2008 so they kept that through around 2013 - then sold the business.

exussum

What would be the expected io rate (mb/s)? I would most likely switch my current storage vps for that.

I currently have a storage vps with inception and understand it's shared hardware but last few weeks the max write speed has been < 10mb/s for hours at time. Then it springs back to life. Assume more storage would be less people per disk and therefore have the disk more often ?

elliotc

@WSS said:
I was going to say it's a bit high, but that's some decent sized spinning rust. I'd probably consider kicking it down to 768M because some asshole is going to try to run Windows on it.

No,plz, it will drive me crazy. Low it to 1/2gb or keep 1gb. 3/4gb make me feel incomplete.

APenguinWeb

@elliotc said:

@WSS said:
I was going to say it's a bit high, but that's some decent sized spinning rust. I'd probably consider kicking it down to 768M because some asshole is going to try to run Windows on it.

No,plz, it will drive me crazy. Low it to 1/2gb or keep 1gb. 3/4gb make me feel incomplete.

703.65 MB it is then!

mikho

@AnthonySmith said:

@WSS said:
I was going to say it's a bit high, but that's some decent sized spinning rust. I'd probably consider kicking it down to 768M because some asshole is going to try to run Windows on it.

That just means 2k3 server instead of a secure (ish) one

Don’t remind of that one customer who still have two running in production

seriesn

@mikho said:

@AnthonySmith said:

@WSS said:
I was going to say it's a bit high, but that's some decent sized spinning rust. I'd probably consider kicking it down to 768M because some asshole is going to try to run Windows on it.

That just means 2k3 server instead of a secure (ish) one

Don’t remind of that one customer who still have two running in production

Who needs ram, when you have swap? Amirit boyz?

WSS

SSD swap is like RAMS!

bikegremlin

For what it's worth:

As a git who's not competent at setting up, running, securing a VPS, I find hetzner box a "better" option - in terms they do all the maintenance. 1 TB storage there is under 10 euros per month, so being a monkey, I'd choose that. Even if you offered half that price, I'd most probably go with Hetzner, unfortunately, at least for the time being.

On the other hand, if you had it configured for FTP access (FPTS, or SFTP), without me having to fiddle with anything (as long as FTP access works) - I'd be happy. No matter how many cores, RAM, as long as it works as storage and isn't too slow.

Though I'm not sure how profitable that would be for you - even if you didn't go much below Hetzner pricing.
I'm also not sure how much in demand that would be - not an expert on that market.

InceptionHosting

@exussum said: I currently have a storage vps with inception and understand it's shared hardware but last few weeks the max write speed has been < 10mb/s for hours at time.

There has been some work done on that over the past few days, it should be improved now.

That said sequential I/O is always going to be low and I cant really pay much attention to that as it is completely expected and there is literally nothing you should be running on a storage VPS where sequential I/O is at all relevant.

From a couple of the nodes atm:

 Timing cached reads:   10760 MB in  1.99 seconds = 5404.11 MB/sec
 Timing buffered disk reads: 304 MB in  3.01 seconds = 101.12 MB/sec
 iops:  avg=819.58

 Timing cached reads:   10802 MB in  1.99 seconds = 5419.97 MB/sec
 Timing buffered disk reads: 628 MB in  3.00 seconds = 209.18 MB/sec
 iops : avg=406.73

seanho

How about an S3-compatible object storage service, with either metered or unmetered traffic?

InceptionHosting

@seanho said:
How about an S3-compatible object storage service, with either metered or unmetered traffic?

So you are saying that is your preference I guess?

To be clear I am not looking to offer a completely different product line, I am looking to tune the current VPS storage offers.

ehab

i would like to see options:

• shared "low" cpu usage for just light backup purposes. like what you might have now. 5€/m
• dedicated or semi dedicated 2 cores - 4GB RAM for example k8s storage worker - nfs medium load- docker register etc .... not over 10€/m
  somehow i feel anything beyond 10€ can might aswell look at the dedicated offers even if its more.

exussum

@AnthonySmith said:
Timing cached reads: 10760 MB in 1.99 seconds = 5404.11 MB/sec
Timing buffered disk reads: 304 MB in 3.01 seconds = 101.12 MB/sec
iops: avg=819.58

Timing cached reads: 10802 MB in 1.99 seconds = 5419.97 MB/sec
Timing buffered disk reads: 628 MB in 3.00 seconds = 209.18 MB/sec
iops : avg=406.73

Looks good I would be interested

InceptionHosting

@ehab said: shared "low" cpu usage for just light backup purposes. like what you might have now. 5€/m

That I can offer.

@ehab said: dedicated or semi dedicated 2 cores - 4GB RAM for example k8s storage worker - nfs medium load- docker register etc .... not over 10€/m

That is not something I can do with that much storage for under €10, honestly for that sort of work load a dedicated server or good VPS with attached storage is better.

Harambe

@AnthonySmith said: That said sequential I/O is always going to be low and I cant really pay much attention to that as it is completely expected and there is literally nothing you should be running on a storage VPS where sequential I/O is at all relevant.

What about backups that you want to run faster than 5-10Mbps? lol.

seriesn

@bikegremlin said: On the other hand, if you had it configured for FTP access (FPTS, or SFTP), without me having to fiddle with anything (as long as FTP access works)

Wouldn't SCP just work too?

InceptionHosting

@Harambe said:

@AnthonySmith said: That said sequential I/O is always going to be low and I cant really pay much attention to that as it is completely expected and there is literally nothing you should be running on a storage VPS where sequential I/O is at all relevant.

What about backups that you want to run faster than 5-10Mbps? lol.

The key word being sequential I/O it is still amazing to me that people still think it has anything to do with real world performance, it is not even a good indicator in most cases.

]# wget http://cachefly.cachefly.net/100mb.test
--2020-05-09 23:35:59--  http://cachefly.cachefly.net/100mb.test
Resolving cachefly.cachefly.net (cachefly.cachefly.net)... 205.234.175.175
Connecting to cachefly.cachefly.net (cachefly.cachefly.net)|205.234.175.175|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 104857600 (100M) [application/octet-stream]
Saving to: â100mb.testâ

100%[================================================>] 104,857,600  111MB/s   in 0.9s

2020-05-09 23:36:00 (111 MB/s) - â100mb.testâ saved [104857600/104857600]

willie

Meh, at this scale I tend to want either super cheap archive storage where I don't care at all about server cpu or iops (serial/network bps still matters), or else I want lots of cpu so I can do stuff locally with the data. A mid-price mid-cpu setup combines the disadvantages of both. Also if high cpu isn't available then being in the same DC can help, so for a while I had a wishosting plan at Hetzner Falkenstein, same location as my Hetzner dedi so I could get to it at gigabit LAN speeds.

seanho

@AnthonySmith said:

@seanho said:
How about an S3-compatible object storage service, with either metered or unmetered traffic?

So you are saying that is your preference I guess?

To be clear I am not looking to offer a completely different product line, I am looking to tune the current VPS storage offers.

No worries, I understand why you'd want to keep it similar to the current storage VPS offerings.

My use cases would be fine with just an S3 interface rather than a full VPS, and I was thinking it might spare you some headache in that you don't have to worry so much about cpu abusers, insecure installs, etc. Just a thought, that's all.

I believe SecureDragon recently started offering ftp/scp accounts at $60/yr for 1TB?

comi

I wouldn't be interested because of raid 5

Harambe

@AnthonySmith said: The key word being sequential I/O it is still amazing to me that people still think it has anything to do with real world performance, it is not even a good indicator in most cases.

Well it does matter for transfers of large files (ie backups) - don't need a 300MB/s seq write but line speed-ish would be nice. It seems a bit better right now (~16MB/s), but I would suggest using a file bigger than the available ram for the VM so you're not just downloading straight into the cache.

The other day I was seeing 30-50MB/s from BuyVM LU and Ramnode NL speedtests dumping to /dev/null - and then writing those to disk, after the initial burst when ram fills up, wget will lock up a bit and it'll jump between 0.5-1MB/s up to say 5-6MB/s of writes to disk.

InceptionHosting

@Harambe said:

@AnthonySmith said: The key word being sequential I/O it is still amazing to me that people still think it has anything to do with real world performance, it is not even a good indicator in most cases.

Well it does matter for transfers of large files (ie backups) - don't need a 300MB/s seq write but line speed-ish would be nice. It seems a bit better right now (~16MB/s), but I would suggest using a file bigger than the available ram for the VM so you're not just downloading straight into the cache.

The other day I was seeing 30-50MB/s from BuyVM LU and Ramnode NL speedtests dumping to /dev/null - and then writing those to disk, after the initial burst when ram fills up, wget will lock up a bit and it'll jump between 0.5-1MB/s up to say 5-6MB/s of writes to disk.

There is a LOT more to it than that though, are you using a multi threaded protocol, what are the limitations of the route between the source and destination, is the transport method using compression, if so what sort of data is it, are you trying to run a 4.x kernel in 512mb ram, are you running your backups on the hour when it is likely that everyone else is hitting the disk/link at the same time you cant simply equate sequential disk IO dd tests to real world performance for backups, it is just not relevant.

Anyway this is not supposed to be a technical troubleshooting thread, the IO is as expected and large downloads/backups are functional, the point I was trying to make is that I personally get a bit tired of synthetic and irrelevant benchmark metrics taking center stage for discussions.

If you are downloading/uploading backups larger than the available free ram you have from a higher latency source and the disk buffers are full due to it being a multi tenant environment and or you are using compression and that is being artificially limited due to you using more than your CPU allowance you are not going to get full speeds.

You can get a cheap dedi with a 1TB for near the same price if you do not like the limitations of shared links, disk resources and don't want to put any thought in to your method (not that i am saying that is the case)

Couple of large speed tests from my VM's on the USA/UK node:

LAX > Phoenix. (11 hops avg latency 13.6ms) - HTTP

# wget http://mirrors-lax.webnx.com/test/10gb.bin
--2020-05-10 02:48:32--  http://mirrors-lax.webnx.com/test/10gb.bin
Resolving mirrors-lax.webnx.com (mirrors-lax.webnx.com)... 104.237.63.187, 2607:f2d8:1:2a::a003
Connecting to mirrors-lax.webnx.com (mirrors-lax.webnx.com)|104.237.63.187|:80... connected.
HTTP request sent, awaiting response... 301 Moved Permanently
Location: https://mirrors-lax.webnx.com/test/10gb.bin [following]
--2020-05-10 02:48:32--  https://mirrors-lax.webnx.com/test/10gb.bin
Connecting to mirrors-lax.webnx.com (mirrors-lax.webnx.com)|104.237.63.187|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 10485760000 (9.8G) [application/octet-stream]
Saving to: â10gb.binâ

100%[====================================================================>] 10,485,760,000  105MB/s   in 2m 0s

2020-05-10 02:50:35 (83.6 MB/s) - â10gb.binâ saved [10485760000/10485760000]

Greece > UK - (13 hops avg latency 42ms) - FTP

# wget ftp://speedtest:[email protected]/test5Gb-a.db
--2020-05-10 10:50:38--  ftp://speedtest:*password*@ftp.otenet.gr/test5Gb-a.db
           => âtest5Gb-a.dbâ
Resolving ftp.otenet.gr (ftp.otenet.gr)... 2a02:587:dff:501::89, 212.205.74.89
Connecting to ftp.otenet.gr (ftp.otenet.gr)|2a02:587:dff:501::89|:21... failed: Connection refused.
Connecting to ftp.otenet.gr (ftp.otenet.gr)|212.205.74.89|:21... connected.
Logging in as speedtest ... Logged in!
==> SYST ... done.    ==> PWD ... done.
==> TYPE I ... done.  ==> CWD not needed.
==> SIZE test5Gb-a.db ... 5368709120
==> PASV ... done.    ==> RETR test5Gb-a.db ... done.
Length: 5368709120 (5.0G) (unauthoritative)

100%[==============================================>] 5,368,709,120 62.6MB/s   in 2m 17s

2020-05-10 10:52:56 (37.4 MB/s) - âtest5Gb-a.dbâ saved [5368709120]

Harambe

@AnthonySmith said: Anyway this is not supposed to be a technical troubleshooting thread, the IO is as expected and large downloads/backups are functional, the point I was trying to make is that I personally get a bit tired of synthetic and irrelevant benchmark metrics taking center stage for discussions.

I agree on synthetic benchmarks, and I apologize for derailing the thread here. I saw someone else mentioning the same issues I was seeing the other day, but as you addressed above those were already resolved. I'm seeing 16-20MB/s sorta thing which is completely reasonable for the price on a 512M storage VPS.

It seemed in the prior responses you were trying to conflate any sequential write speeds with synthetic benches, but there are a lot of legitimate use cases for large sequential writes on a backup VM.

---

Back on topic: I think the pricing is fair for 1.2TB of storage on a 1GB KVM. A 'nice to have' would be a separate boot partition of say 10GB on an SSD, so the core install and random read/write tasks would mostly hit the SSDs and then you were free to partition up the space however you'd like.

I'm not sure how doable that is with SolusVM/whatever panel you're planning to use, but I think it would be a nice feature and would free up iops on the spinning rust.

InceptionHosting

@Harambe said: I'm not sure how doable that is with SolusVM

Not without manual work being required for every provisioning sequence.

Also I don't think I can think of a single unbuffered sequential write case for a backup, but I think I will make a new topic about all this.

WSS

Technically, one could provision a small OS container and then setup LVM for the disk image on the spinnyrust. It wouldn't be all that difficult to hook, but eh.. not sure how much I'd care, either. Actually seems something more in line for OVZ of all things.

Ympker

@bikegremlin said:
For what it's worth:

As a git who's not competent at setting up, running, securing a VPS, I find hetzner box a "better" option - in terms they do all the maintenance. 1 TB storage there is under 10 euros per month, so being a monkey, I'd choose that. Even if you offered half that price, I'd most probably go with Hetzner, unfortunately, at least for the time being.

On the other hand, if you had it configured for FTP access (FPTS, or SFTP), without me having to fiddle with anything (as long as FTP access works) - I'd be happy. No matter how many cores, RAM, as long as it works as storage and isn't too slow.

Though I'm not sure how profitable that would be for you - even if you didn't go much below Hetzner pricing.
I'm also not sure how much in demand that would be - not an expert on that market.

Isn't SFTP enabled on most Ubuntu/Debian VPS by default though (in combination with open ssh)? Ofc, securing a vps is another story. But the SFTP part should be easy.

beagle

@Harambe said: A 'nice to have' would be a separate boot partition of say 10GB on an SSD, so the core install and random read/write tasks would mostly hit the SSDs and then you were free to partition up the space however you'd like.

LetBox gives your server provisioned on a NVMe drive and your storage drive is provided as a raw drive (it's a block storage attached to the VM). You can then partition it and format it as you wish.

UltraVPS also provide the OS separated from storage on storage servers. I find it particularly useful because you have your full allocation for storage. For instance, their 500G storage product gives us 10GB for OS and the full 500GB for storage.

Not sure how many tickets you may have with users not knowing how to partition and format the storage though.