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.
Comments
hey just wondering are the ports shared? what is total rack capacity(top of rack switch/router)
Not shared.
Typically 20Gbps for 40-48 nodes. We yet to have see even 5Gbps 95th percentile exceeded.
New DC plan is to have 10Gbps per 8, aggregated to probably 2x100G uplink from the rack itself.
Meaning 336 sharing 100 or 200Gbps.
We haven't finalized the spec.
Good stuff, any plans to add optional 10gbit port to future setups
Hardware always gets better over time, so certainly eventually. No idea when is it. Might be 10 years from now.
Just be mindful, 10Gbit is like an instant +200€ per month right now
New Models AND New Stock!
There's a lot of MD units available right now;
Prices coming down quite a bit :O
Take benefit before they start rising again
Units going fast, but no worries, more is always coming
For some price low was reached and pricing is again increasing.
rdns when?
After other basic automation has been written.
No specific ETA. This will take time, actual time, not like "by tomorrow" but more like Q3/Q4 time.
Turns out, we have to code everything from scratch ourselves. the most promising automation for these has "single server mindset", when we need to think about "1000s of servers". Just setting up single server on that system took more than 10minutes and it's built in fashion it demands 100% of your attention.
Further, maybe the biggest issue was; It would've now costed 0.50€/Mo/Server sure, but how about in 10 years? They've already said price is going to increase. What if a investment firm buys them out? 2€? 5€? 10€/Server/Mo? How are you going to sell 10€/Mo dedis if your licensing is 20-100% of the revenue?
Also we want to streamline the actual OS installation, Stress testing etc. we've not seen this done on anything available commercially
Let's see where we wind up, it's too early to tell what direction we finally end up with.
Considering our seedbox solution is 100% built in-house, and is the only one which has full end-to-end automation, we already have a track record of delivering stuff like this
Compared to coding that for SBs, this is easy. Very very easy.
Dang! Some of these new 4TB Nodes are insanely fast:
This is one of our new MD20 nodes, now sold out. Customer paying ~32€ per month for this. i5-6500t/16G/4TB NVMe
The "slower models" still faster than joe average nvme;
Current pricing:
Prices have mostly stabilized now with slow rate of change right now;
No AMD Dedicated Server?
By the way,how long do you need to handle a ticket on average?
Depends, I guess - setting rdns (manually) was denied, even when paid for 3 years term for dedi. I am still super pissed about this.
AMD units are coming, some 2400GE level at first only tho.
Depends, sometimes ~60minutes, sometimes a lot longer. It also depends on your definition of handled. Resolution or just answer?
Btw, one should expect anything under 8hrs at this price class. Large companies you have to pay extra for 8hr response times. Response, not resolution. Large hosting providers typically only offer support 8/5 not 24/7.
That's the keyword here. 10x or 30x the price and we can start doing things manually ... maybe ...
There's a reason why companies charge for "free stuff" -> It's the documentation, upkeep and most importantly; Remembering to clean it up when it's not needed anymore.
So in other words; If we'd handle stuff like this manually, you wouldn't have the server to begin with.
It'll be available when it's automated. ETA i do not have.
There was a bug in dynamic pricing which resulted in new models having too low pricing, this was fixed, along with some other algorithm changes to better reflect the hw config.
We received a new shipment today, but it'll be a while before those are online.
Steadily stock dwindless, prices are mostly stable tho;
Let us know if there's a specific config you'd like us to see add next (i know i know N100; Those are coming!)
This is where we will house up to 9300 of these nodes over the coming years ... just a few steps remain, not too many
HDDs ex 2x4, 2x8..but I'm guessing there is no room?(mini PCs?)
I currently have a md and I’m pretty satisfied with it, but I’d love to see a 2x4000 nvme config.
You can attach a storage box which works kinda okay. I can't saturate 1Gbit transferring a file but that might be user error
actually .... yea no, not currently.
We have an idea how to make that happen, but it's a bit ways off.
you read my mind! I've been wanting to see that too, just worried there's no profit in those for us. 2x4TB model would cost multiples of say 1x1TB model. Esp. with 2x high performance drives, seeing insane performance from just 1x high perf 4TB, would be wicked to see dual.
Yea that's a SINGLE NVMe on one of our 4TB models ...
It's outpacing a 2x1TB NVMe model i tested it against. We are just starting to build our database of real life performance of various nvme models
Might make them reality just because we can tho
Indeed, and if you mount using rclone it's trivial to use the local NVMe as cache for it.
Performance wise, i don't have much more info than that
Waiting for more real world use case reports like yours
Thanks!
Yep, rclone and sshfs I tried. I can get around 650Mbit which is fine really, although I did expect to be able to hit gigabit as the machines are in the same DC. I suspect the storage box hits its memory limit preventing full speed but I don't have anything solid to back that up (mostly just seeing the memory limit get hit when doing my testing of various parameters / methods of moving data). You'd think 500MB would be enough, maybe it is...
10gbit would be cool
yea the storage box might be just busy. That's where the caching steps in
You have to config cache separately with rclone
Indeed that would be, but that's a few steps away still. 2.5Gbit we will see sooner i think. Not sure.
10Gbit also adds like 100-200€ in cost, not sure much but it's a lot. Bandwidth is really expensive.
Part of the reason is that just adding 10G to a node might triple it's hardware cost as well :O
@PulsedMedia would you be able to offer servers with GPUs for machine learning? At work we use Google Cloud for everything, including this, and GPUs are quite expensive there, so I am starting to thing that I should probably look into alternatives. We are based in Helsinki too.
Just as a reminder, few lucky ones get THE BEST BRAND RAM with these units ;D
Hmm, we got plenty of GPUs up to RX6000 / RTX3000 in stock, but we would have to charge arm and leg (relatively) for setting up servers around them on long contracts. Otherwise, not a problem.
Pricing is based on presumption of 100% usage 24/7. When rented people don't care about power consumption either, so not power saving max efficiency like miners have either, the primary OpEx on these being the electricity.
Let us know the config you need, and min contract is 1year to get it sensibly priced.
Some units back in stock now.
At least a reply.Of course not 24/7 support needed.It’s OK for me if you tell me that you are busy and let me wait for a whole month.
NO SETUP FEES FLASH SALE -- FOR 14HRS
Setup fees have been disabled for the next ~14hrs.
So pick your node now;