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.
HK Storage VPS
randvegeta
Member, Host Rep
in General
Will be offering HK based storage VPS very soon (Early October is planned).
Before we finalise the hardware, just wondering how existing storage VPS performs with regards to I/O. And more importantly, how much is actually needed?
Our plan is to run a ISCSI storage nodes with RAID 5 or RAID 6. Performance is obviously not going to be great compared to RAID 10, but of course you get much more disk space for your buck.
100-200 IOPS and 30-50MB/s is what we have in mind.
Opinions welcome.
Thanked by 1shc
Comments
You meant early November?
$3/month for 50GB, i'll take it.
Whenever a LET provider says 'raid5 or raid6', I parse it as RAID5.
I hope I'm wrong.
Will it have an IPv4?
Yes November. Sorry
Raid6 is more likely depending on the number of drives we can squeeze in the box.
Yes it will be ipv4.
I was thinking of much larger space for that kind of price.
5 USD per 1TB ? that's what I have in mind for LET price , not really sure if it's possible on HK though...
6.66 USD for 666 GB.
That's a bit of a stretch to say the least.
I'll do that deal!
Interestingly, that's actually not far off from what I had in mind.
If the performance was limited to say 100 IOPS and 30MB/s, RAID5 is absolutely doable. RAID6 may still be doable. 8 drives per storage node would mean 25% overhead, which is a fair bit to lose on a budget storage VM.
Still.. I was originally thinking more along the lines of 50GB per $1. 666GB for $6.66 is half the price....
Without BW & with RAID6(0)... extremely hard, the DC space and especially power w/ AC is very expensive in HK. 5$ is hardly even doable in Europe.
You'll like your RAID5... until it dies. Dangerous especially in software (RAID6 is obv also, but way less).
its possible, but with 300GB bandwidth. storage is not expensive, bandwidth is in HK.
Hello there, do you have a moment to talk about Jesus Christ
RAID 6 it is!
The biggests cost will be bandwidth. The power should be minimal. It really depends how dense it can be done and how much of a performance restriction people can accept.
The potential problems I see are when all users are doing dd at the same time, or trying to run some io heavey software.
But for actual storage, I dont see why anyone would ever really need any more IO than the bandwidth they are allowed to use.
Useless, you need to compare RAID5 to RAID6 AND a 3 parity drive setup - plus if using ZFS you do not have bit errors at all and faster rebuild.
My comment also specifically refers to Linux MDADM RAID5/6 as being horrible af, while general HW RAID5/6 is just crap but not that bad.
Full disclosure then. The storage servers will be using MDADM software RAID (or possibly onboard motherboard RAID...) and the storage servers will have 0 backups.
Bandwidth will be pretty much pure HE.net + HKIX.
And as mentioned above, $1 per 50GB storage per month was the target with 100GB data transfer. Ergo, $7 /month = 350GB space + 700GB data transfer.
Though I like the idea of 666GB space for $6.66 with 666GB monthly data transfer.... A fiendish idea!
Not much horrible about it in this particular aspect. On the contrary with software RAID you have more freedom and flexibility to salvage your RAID5 even if another drive has nonfatal issues during a rebuild (e.g. a few unreadable sectors), whereas a hardware controller is likely to just fall on its back, throw its hands (and legs) up in the air and lose you the entire array with no way to bring it back.
Full YOLO?
Jesus Christ is not useless.
I'm unsure if it's bait or not. Hopefully, it's bait and I've fallen for ye olde ZFS pitch just a little bit.
You don't need to hamper performance & lose available storage with an additional third parity and you can't mistake a RAID for a backup.
Pretty much this; albeit I'd use a HW RAID for a professional setup, assuming the node is heavily loaded.
Only way to fly
After much consideration, the most cost effective option would seem to be no RAID at all.
Current plan is to have a storage node with regular disks. 1 volume per disk. And a live backup storage node with identical configuration, mirroring the first node, and have them setup in HA. Kind of RAID 1 striped over 2 physical storage servers.
Reliability and performance should be superior to RAID 5/6 and actually, starting out with our existing stock of unpopular (slower, older...) disks, more cost effective (at least starting out).
Testing underway. Release date may be next week.
Not using raid has its merits.
Though I do use raid for production, I absolutely refuse to use raid on my personal computers. I just use JBOD on my storage PC at home with just backing up absolutely necessary files.
These storage VPS will be cheap and are intended to be used for backup purposes.
With the inclusion of the live backup in HA, I see no real reason to have RAID on the node. Triple redundancy?
You could use zfs send (staggered pve-zsync per 15min)for the disk mirroring part if you're going with proxmox.
I think this is Low end innovation in storage type.
Not using Proxmox. Going for a XenServer + Virtualizor + iSCSI Storage.
It's been built and is now undergoing some testing. Will start with a small.. 10TB cluster, and see how it goes in production given the I/O limits.