Indeed, NS and SL look like similar offers (maybe not apples to oranges, but perhaps tangerines to oranges).
Here's a SL offer similar to the NS Medium offer:
SL Bare Metal:
8 x 2.0 GHz Core Bare Metal Instance
8 GB RAM
1000 Mbps Public & Private Networks uplink
2000 GB Bandwidth (outbound; inbound is unmetered)
250GB SATA II
$319/mo.
NS Medium:
2 x Xeon 3.2Ghz
2GB RAM
(not sure of uplink; I don't need more than 10Mbps)
2 x 73GB SCSI Raid 1
$0.17/hr or $122.40/mo
+ $28.80/mo. for 8GB RAM
$151.20/mo.
NS is much cheaper.
NS has less HDD space than SL (73GB vs 250GB).
NS has fewer cores than SL (2 vs 8).
NS has a free shared HLB when you get >= 2 servers (not sure about SL).
The SL offer is more quickly deployed; since NS' offer requires an add-on, it would take more time to deploy.
---
Perhaps comparing the above SL offer to a NS Jumbo (below) makes more sense.
NS Jumbo
2 x E5405 Quad Core 2Ghz
8GB RAM
(not sure of uplink; I don't need more than 10Mbps)
2 x 500GB SATA RAID 1
$0.38/hr or $273.60/mo
This supposedly can be deployed without a day or more delay.
---
What are your thoughts on SL in general? They are certainly more well known. I'm not sure how much that should factor into one's decision.
---
I wonder if anyone cares to comment on the "Bare Metal Cloud" vs the "Public Cloud" offers by SL.
The chief difference seems to be storage options. The "Public Cloud" uses SAN while the "Bare Metal Cloud" has its own local HDD. I'm no expert on SAN vs HDD. Has anyone found any benchmarks (IOPS, latency) for these two SL offer types?
Heh, I guess the point of these instant provisioning things is to you know, get one of each for an hour and try it myself! Newman!
I was wondering the exact same thing. I mean, someone who offers dedicated servers surely should have a responsive website. I wasn't going to say anything because I thought it might just be me - but I guess not.
That is the datacenter our company uses. Cat 5 rated, plenty of power, hosts a slew of EMS and other core services. Some parts of it aren't the best but you could do a whole lot worse in Florida.
What intrigues me about this is the possibility of leveraging cheap clouds that are literally racks away. Hmm...
I don't understand it because their content is very Buzzword Compliant, but this is what I can gather: they sell you dedicated servers billed hourly and provisioned quickly, much like a cloud provider.
It's a cloud provider that doesn't use Virtualization. You don't need Xen or OpenVZ to have a handful of boxes and an API to address them as a distributed redundant machines. (Does anyone even know what Beowulf clusters are anymore?)
That's a pretty cool setup, really. I almost made an ass of myself by saying "they should be using blade servers" when they are, in fact, using blade servers.
I would really, really like to know how the power density of blades like this compares to the potential power density of hefty machines running virtual instances. If you can get close with dedicated blades, something like this would make a really desirable (and sustainable) premium "cloud" offering.
I became interested in NewServers while looking for affordable dedicated server hosting providers. I am personally less interested in "cloud" scalability for my pet project, and more in getting a "decent" quality dedicated server or 3 for a "decent" price (lots of hand-waving here I know). That is, I find EC2 to be "expensive" for what I want; I'm not super profitable yet in my tiny little startup so saving money yet maximizing server performance at the same time is the tradeoff I'm researching.
I have not found any hosting providers besides NewServers that are as affordable for "lots" of RAM. Also, NewServers' bandwidth quota is 3GB/hr with $0.10/GB overage.
NewServers offers add-ons like extra RAM. These require maybe a couple of days but for me, I don't need pure cloud scalability in minutes. I just want a few cheap dedicated servers.
Consider this configuration;
Intel Processor Cores RAM HDD Price/Hour
Jumbo 2 x E5405 Quad Core 2Ghz 8 8GB 2 x 500GB SATA RAID 1 $0.38
An always-on Jumbo is thus:
Jumbo = 720*0.38 = $273.60/month
RAM add-ons:
8GB $0.04/hr 720*0.04 = $28.80/mo. more
16GB $0.08/hr 720*0.08 = $57.60/mo. more
24GB $0.12/hr 720*0.12 = $86.40/mo. more
32GB $0.16/hr 720*0.16 = $115.20/mo. more
Thus, a Jumbo with 24GB RAM = $273.60 + $86.40 = $360.0/month with lots of bandwidth transfer.
Compared with something like EC2, this is a no-brainer on the cost front, provided you do not need the other Amazon features... (I don't, but that should be made clear).
Also, I talked with Alex at NewServers and their turnaround time for provisioning servers with the above add-ons is "2 or 3 days". Good? Bad? That's fine _for me_ but do note that servers with add-ons are not provisioned on the order of minutes but days.
Thoughts? What am I not considering but should be?
For example I rent a few UK based dedicated servers which cost me (converting in my head here) $150-$190 a month. Specs are similar to their Fast package but with 8GB of RAM and unlimited bandwidth.
That is around average price here in the UK; I assumed things were similar in the states, but I guess not?
Well, maybe there is some room for this in the market. If they can really provide instant provisioning of a dedicated server, then this should be more efficient than a virtual machine based approach.
Their API looks pretty decent, so you could use this in a similar manner to EC2. The downside is that instead of things like S3 storage, you get iSCSI. Sure, it's dedicated, but I don't know if it's redundant. And if you want any other cloud-like infrastructure (Queues, Simple DB, etc...), you're going to have to build it yourself.
While you could get more power out of a setup like this, I don't see how it can scale. I mean, we're talking about a dedicated machine for each user. And this is also for each class of machines. They'd have to over-buy machines by 10-20% to match unexpected demand. And this is for all classes of servers (and they have 5). What happens if they don't have a machine of your class available? At least with a virtual machine approach, you can buy big servers and slice them up in a variety of ways.
We do not keep any servers unused, all servers are recycled
and added to queue so the customers have equal rights to get
them... and we cannot keep them reserved because of the costs...
instead you should consider leasing them for longer periods
so you have them available.
I interviewed these guys at Structure 2008 for my blog. They were pretty nice and rather reasonable people. They seemed to have an understanding of their strengths and weaknesses. To be fair, they're not targeted at the Web 2.0 market that EC2 has targeted. They're more suited as an alternative to dedicated servers, and seemed to be targeting traditional businesses who are bad at capacity planning.
Could you elaborate on your comment about targeting companies bad at capacity planning? I did not get that from the interview at least. Also, I am considering using them and wonder how I might be implicitly bad at capacity planning :)
Forgive me, I'm a much better technologist than I am a blogger .. Hence the sparse number of entries on my blog!
In general, I feel that the whole EC2/instant provisioning model of servers, which for any moderate (10+ serverS) application tends to be far more expensive than owning your own hardware over the long run, is a sign of an infrastructure that either does not excel at capacity planning, or that is unable to plan for capacity.
