Ask HN: Recommend a Managed Hosting or Colocation Provider? Our company has run into performance bottlenecks with our current cloud provider (we're very I/O-centric). We're primarily interested in Managed Hosting, possibly with a hybrid cloud solution. We're looking at: Softlayer ThePlanet LiquidWeb/Storm SunGuard Have any advice about these companies or any others that make compelling Managed Hosting partners? |
I understand you're set on a cloud-like environment but have you considered running off a dedicated server(s) with SSD? The IO improvement is dramatic. We have several clients running SSD in production without any issues to report.
Regards
Joe
We have a combination of random and sequential data access patterns. Our data fits really well into key/value data stores, such as Cassandra. We'd prefer to get 10k qps for random I/O, of course that's data-model dependent.
Our Data is currently in the hundreds of gigs to several terabytes, but is likely to increase quite a bit in the coming months.
If we buy hardware, we're definitely planning on testing SSD extensively.
The main question is about who to Colo or Rent from.
Keep in mind that writes will always be faster than reads thanks to cache write-back. This is because writes are asynchronous at the kernel and controller level. When your app writes a file, it indirectly writes it to the disk cache. If the data is needed almost immediately after the write, chances are the read is going to be as fast because it's still present in the cache.
Now, the reason why reads are generally slower than writes has to do with the fact that the app must wait for the read to complete (to present the data to the user). So be sure to focus on read performance and get a good UPS.
I should have asked for a budget ballpark. But let's assume you can't afford buying several 250GB SSD disk (double for RAID-1). That's where I would recommend a stripped mix of SSD and SATA RE3. So if you have, say, 2x250GB SSD and 2x1TB SATA, you would RAID-1 the SSD pair and RAID-1 the SATA pair then +0 RAID across the resultant logical volumes. That's RAID-10 but with manual stripping. You can save quite a bit this way. Also, be sure to choose a small tripe size so each file is split amongst the RAID-10 evenly and improve performance (ex: 40% of the file is on SSD and 60% on SATA).
I'm not sure what to recommend as far as hosting because I don't have any info here. Colo has its advantages but it can get expensive. Colo costs can be excessive. The hardware, staffing, electricity, bandwidth, routing/switching equipment, maintenance, etc. Be sure to colo as close as possible to your office/home so you can take care of emergencies quicker. DC staff isn't always to the most competent so you'll either have to learn as you go or hire someone.
Renting is interesting (let's say you're getting a fully managed node). You don't have to worry about staffing, 4AM trips to the DC, contracts, hardware depreciation, replacement contracts, or contingency plans when the main server decides it won't boot. It's all handled by the provider. If you pick a smaller provider, you'll be sure to get better service as they tend to have more of a personal service.
Renting a multi-TiB dual quad core server with SSD and SATA can be anywhere between $800 and $1200 per month depending on server management and bandwidth. Colo costs are more complex and depend on many factors.
Keep in mind that I can be biased considering my role / occupation in the managed server sphere.
Regards
Joe
http://www.datacenterknowledge.com/archives/2008/06/01/explo...