Apps and Services

[houltmac]

I have been dealing a lot with issues surrounding server and cloud services taking over from local services lately. Starting with my testing of OS X server, the whole issue is growing in my mind, and I am not getting too far in finding a logical solution.

Take iCal server in Mac OS X 10.5 Leopard server for example. This works great over a local network, and users have control over who can do what with their calendar, how often it updates and so on. Still, the issue of all that data being stored on a server is undeniable. There are multiple points of failure here, starting with the data itself. If that server (or in our case, RAID) dies and there isn't a backup then you are stuck. If the RAID gets too big you can't back it up offsite. This is obvious, but what about the network.

We run a "gigabit" network at work which is great. The server only has one ethernet connection however. What if the port dies? What if the cable has issues? What if the switch dies? These are all single point of failure which could bring down the entire company, not just calendars.

Running a little deeper with the issue then, what about cloud services? Connection speed is obviously a much bigger issue here. While even our wireless (802.11x) networks are getting very fast, most DSL and cable connections are still painfully slow which limits their usefulness. Upload speeds are an even bigger concern than downloads. While downloads are averaging what, 4-6Mb (realistically 1-2Mb during business hours around here) we are still measuring upload in Kb. This isn't so great for backups, let alone actually working on a document to save to the cloud. There has to be a way to cheaply compensate for this failing, or just plain up the bandwidth.

Again, you also have single points of failure in the line, the router and so on. Multiple lines can help, but buying different lines from different ISPs, running off different exchanges is expensive, administratively painful and all round undesirable. This is still the only way to get close to getting genuinely likely uptime stats above 9x% as far as I can see.

Of course you can use failover and load-balancing to make use of the additional lines, and this is another thing we have been testing recently. We have the ability to utilize our "backup" ADSL line for only certain kinds of traffic (FTP, POP, SMTP and ARD at this point I believe). This is very cool, and effectively widens our bandwidth, making the second line much less difficult to explain to the accountant (although she still won't like having 2 ADSL lines as well as 2 SDSL lines but hey). The really cool thing about our setup is that if a line fails it's auto-detected, we are emailed, and all traffic is re-routed down the other line until the issue is repaired. Very cool; but not perfect. If we are in the middle of a huge FTP transfer and all the data is failed over, it disrupts the connection for up to 90 seconds and the transfer fails. Darn.

This isn't to say I don't think that server/cloud services aren't great. The idea that we can so easily share data, present ideas and so on without actually sitting in front of one machine is fantastic. I especially like the idea of cloud services with local applications; it could be the future even. As far as I can see this gives the best of both worlds: safe, secure and accessible data services with the rich environment of desktop applications. I love Flickr for sharing and storing my photos, but I also love to edit them in iPhoto/Apertire/Photoshop. I love being able to watch video on Viddler or Youtube, but couldn't imagine trying to edit outside of iMovie/Final Cut. This seems to be a great idea to me except for one point; it's totally reliant on my connection.

If my connection is slow, or worse - inaccessible, what am I to do? While ubiquitous networking is a bit of a rant right now in the industry we are still a way off, and until I can access the cloud on a plane and so on it's always going to be a pain. If any single point of failure succumbs to the worst, we would be stuck. What then is the answer?

[Yuiichi]

I think the answer to this is that we're trying to make these cloud apps too fast. Sure, Google Docs is great and everything, but it's also very new. Not time tested. You could lose or not have access to stuff in certain situations. There will always be things that plain, offline desktop computing will be better for, such as video processing apps. Can you imagine trying to edit a large, HD video over our current bandwidth? I can't. You did address this, so I won't go further into it.

I guess what I'm saying is that we need to give these apps a few years to mature and fully realize their potential. As you said, a single point of failure really breaks the whole deal, which is something that has to be worked around. In my mind, I could see that point of failure (which would be disconnection from the network) be solved through having a corresponding offline app on your local machine. Although this is very Web1.0 thinking, I don't think that there is really a better solution to be offered. At least I don't have it.

Interesting post Matt, although I'm always a little weary of these cloud type apps!