Dropbox is terrible with CPU usage. This is something I reported to them several times, but it seems they don't really care.
Last I tested, Dropbox consumed CPU for ANY changes ANYWHERE in your filesystem, not just within the Dropbox folder. So its (slow) code processed changes even when you untarred a large set of files somewhere in /tmp.
I also stopped using Dropbox for storing my code because of that — it's nearly useless, as it consumes so much CPU and eats so much battery life, that it simply isn't worth the effort.
Another problem with Dropbox is that upon login it indexes the entire Dropbox, which if you have an HDD basically kills the machine for several minutes, e.g. you can't do anything else. That happens to me on a Mac Pro. On an SSD-equipped MacBook things are better, but it's still a monstrous operation.
And to everyone that says you should not store code in dropbox: a) there are good and legitimate reasons for doing so, b) dropbox could work a little more on optimizing CPU usage instead of adding "save your photos to Dropbox" features.
I do use git. But I often develop on two machines: a laptop and a Mac Pro. Which means I am often in the middle of something when I have to just pick up the laptop and go. "In the middle of something" means that just pushing commits over doesn't cut it. I can (and do) rsync stuff over, but that is annoying and very error-prone (it's too easy to sync stuff the other way around and lose work).
I guess you could say Dropbox gives me finer sync granularity than git does.
Dropbox seems like a perfect solution: have a single place where you keep code, develop on two machines, just pick up your laptop and leave whenever you want, without worrying about sync.
I don't understand. You can easily cut a branch commit. Push the branch. Pull the branch finish it. Commit. (optionally rebase). Delete the temporary branch. Git handles this problem very well.
Right. I get a call, I need to leave for a meeting in 5 minutes. This is exactly the time when I want to cut branch commits, push branches, pull branches, commit and rebase. Even rsync is better than that.
Put it another way, I don't understand why you'd use Dropbox at all, after all you could easily version all your files in git, using push/pull/branch/commit/rebase to get your data to your different machines. Git handles this problem very well, after all.
The specific problem with git in Dropbox is that working in git can cause a large number of transient files changes as you, for example, switch branches. Dropbox doesn't know those are transient changes and attempts to sync all of them. That is one reason it can drain battery so fast.
Obviously with regular files, that is not an issue, so Dropbox works well. (unless you're managing them with git, I guess)
If you're willing to accept the "lightness" of git branching and commits, it's possible to "save and sync" your work very quickly. Assuming you're on a dev branch:
Then when you're ready to work on a different machine, you fetch from origin.
If you want to keep your commits meaningful on your dev branch, you can create a temp branch first, then rebase later to get rid of the ugly temp commit.
I don't use dropbox at all? To me it is a great solution for consumers but my needs are different. I use git + standard unix tools. For backup I use tarsnap. It rocks.
You don't understand or you didn't spend ti e thinking about that. It's doing all those steps, or doing nothing and achieving the same result. Not sure what you don't understand.
My use case: The "extras" (files not in the DVCS like libraries, dependencies, etc) are just as important for working, but they take too long to transfer/update from one PC to the other, especially for things like GAE which don't have easy ways to manage.
I use both dropbox and git for my home projects. I sometimes develop on my desktop, sometimes on my wifes laptop, and sometimes remotely on another server through ssh, and I don't want to commit/add stuff to the git index just because I move from one computer to another. I could do this with branches, but it's too much hassle when I just want to shut down computer.
So I work on a directory that is in dropbox, and in git, and I can instantly shutdown my PC and start writing code on the laptop. Commiting changes is a separate step. It's quite convenient. Also I can test the code by accessing dropbox public link (it's mostly javascript and static websites), and there's no separate "deploy" step - when I save a file on any computer the change is ready to be tested, always at the same place. With git I would need to remember to commit, and update before I test, or I would need to set up a test server on every computer, and depending on where I work I would need to access different address. The difference between 15 seconds and 0 seconds to test small change is VERY significant.
As opposed to there probably aren't really good reasons.
Doing it in addition to source control seems like something everyone should do though.
The way people generally use source control isn't to push every single line added or changed to a remote copy of the repo as soon as it is done. A commit is more tied to completing a task, you need something else to cover you for loss for smaller amounts of code than that if your computer dies.
More specifically to me, it depends on what branch I'm on. If I'm on dev, I have no problem with big commits. (For features I'll often set a bookmark [hg]- maybe even clone and work from the clone if it's risky).
But if I'm on test or hotfixing on stable, I use very small commits so that I can easily graft (backport) back to dev.
But at the very least I have more than one commit message like -FIX blah blah commit so can get from <name of other machine>
My entire work folder is inside my Dropbox. Most of those projects are under Git version control, other temporary folders, toy projects and bits and bobs are not. This way my work folder is nominally backed up and synced across my machines. I wouldn't advocate using Dropbox as the only/primary backup and version control for important projects, but its totally legitimate to have code inside of a Dropbox.
I use dropbox as a client facing equivalent of FTP.
That way the client can have programmers/designers change their UI without needing any access into the server. It also allows me to share each site and doesn't bring the entire headache of managing accounts. I simply share the yoursite.com folder to the client. If a client has multiple sites it's just as easy to share all of them.
same question here. If you can get free private git repositories on services like bitbucket, it's sort of a mystery to me why one would want to lose true versioning and go with plain backup instead.
You can have both. If you put your working directory in Dropbox you will have a copy of your uncommitted work as well.
Also, in many cases it is easier for a graphic designer or just a customer to put stuff in a Dropbox folder that will sync to their website, than uploading stuff using a web form or learning to use git and X amount of deployment tools.
I'm part of the camp that uses both Dropbox / Git for my development work. I've actually had Dropbox save me a few times due to nasty rebases. For a lot of one off scripts, I don't really commit them into git so that's where it comes in handy. The extra redundancy doesn't hurt...
One advantage is near-instant sync between many computers. With a VCS you have to push and pull manually or set up some complicated cron jobs to automate it.
That sort of glitch sounds like the kind of thing Dropbox does (did?) from time to time because it's not aware of bundles. (For those unaware, on OS X many files are actually folders, eg: http://en.wikipedia.org/wiki/Rich_Text_Format_Directory)
> Another problem with Dropbox is that upon login it indexes the entire Dropbox
I haven't used Dropbox in a while as a local client for other reasons but I think people need to appreciate this is the payoff for having the convenience? How can one expect it to sync changes if it never scans or detects changes? People seem to want a magical product where it never uses any HD throughput or CPU to index/scan/upload megabytes of data, sure it can look for changes less often but that decreases the usefulness of the product.
As a sidenote, surely all the dropboxers use dropbox for everything (they have unlimited accounts) so they must be 100% aware of this? in which case there must be concrete reasons for this!?
I am not saying Dropbox sucks, I am just criticizing those aspects which I think could be improved and are being neglected.
From my conversations with Dropbox support I understood that Dropbox really can't handle a large number of small files. The large plans are intended for people who store large files, not lots of stuff.
The indexation doesn't need to happen immediately upon boot. Do one directory every minute or so, let the computer start up and do other things while your background task runs.
Changes are detected because the operating system notifies dropbox of changes made, they're not continuously scanning the drive. From other comments it sounds like dropbox has registered a hook like that for the entire filesystem, instead of just the working folder. That seems unnecessary.
My computer comes to a halt on boot because of the Dropbox client. I hate it.
For some reason SugarSync does not do this kind of thing, so there has to be a way to do it in a smarter way.
Why can't Dropbox remember the state of the files in your dropbox folder, and upon login, grab state of the files on the Dropbox server and do a simple comparison, without re-calculating hashes for the files on your computer?
A hashing/checksumming filesystem would make it possible to determine whether a file's contents have changed without reading the file. The burden of computing the checksum is then shifted to writing, which in theory occurs once per sync in the case of Dropbox. This burden is merely computational, rather than a complete disk I/O burden imposed by hashing the file when Dropbox restarts.
The new battery indicator on the menu bar identifies apps that consume the most energy, and Activity Monitor now details energy consumption per process.
That finding might be quite interesting and all... but I'm thoroughly confused about that guy always quoting power consumption in mAh (which is the unit for electrica charge).
I'd also check if that guy's dropbox daemon(s) are stuck in an obvious loop. I'm logged on in my private server, and when it's idle there's hardly any activity on the dropbox threads, the main CPU consumer is one servicing what I suspect to be the main select() loop. And that thread basically stat()ing the .dropbox/config.dbx file every 2 seconds, and reads 16 bytes out of it. That should hardly be measureable in terms of power consumptions on a modern CPU.
No, that is not a typo. Thos is a typo - an error created during the mechanical act of typing. Yours was just a plain mistake, you got the units wrong. Own it and learn from the error instead of excusing it away.
I didn't see the article before fixes, but I seem to often type "driver" instead of "drive". If one typed "mAh" enough, don't you think a similar thing could happen?
i actually do the same as the op and i really like to have a seperate backend up copy of the code in dropbox BEFORE i commit to git. Furthermore i dont really want to commit only because i am changing machines, but if i have something significant and non breaking to commit. This is less of a problem with git as you can just work in your own branches very easily, but still, working on code on multiple machines is just seamless with dropbox.
I store a lot more than just a git repo in Dropbox. I also use it to share files, etc. But yes, I'm starting to look at alternatives. Maybe I'll develop a small utility to push/pull/share files from my home NAS upon request rather than having a continuous process running in the background.
You can also, what I did, switch dropbox off and switch it on only when you are connected to power for a sync. It's not ideal, but it works. I don't see how dropbox could do better though. If you do it manually to your NAS, you basically do the same as switching dropbox on/off or dropbox in a cronjob. Anything monitoring your file system and synching up with many small files will incur load like this.
I use ControlPlane - http://www.controlplaneapp.com/ - which monitors your "context" (like location etc.) and lets you trigger events on context change.
Just added some rules so that Dropbox quits when I'm in my "mobile" context, but reopens when I am in my "home" or "office" contexts. Will see whether it works on my commute later.
I had this happen in the past too and found it to be the result of two issues:
1) Using dropbox to house a very large git repo with massive changes and tons of branches... Every time I switched branches, the hundreds of files would change and obviously kick off a major sync job.
2) There were a few files in a folder than had the owner set to the wrong owner and Dropbox couldn't get permission to access them. This caused it to get stuck in a loop trying to sync those files and Dropbox ended up running my CPU up to over 100% load.
Fix those issues and you shouldn't see it causing any more problems with your battery life.
Try this: separate out the git directory from the working directory so that the working directory is not in dropbox but the git directory is. That might solve some of your problems.
(1) you can also use the GIT_DIR environment variable, which might save some typing during development
(2) another solution is to just clone the bare repository in Dropbox to a temp-directory, work there, and push the changes back to the Dropbox'ed repo. Saves you the effort to specify GIT_DIR or --git-dir all the time and is the saner approach if you have to switch back and forth between different repos.
> Using dropbox to house a very large git repo with massive changes and tons of branches... Every time I switched branches, the hundreds of files would change and obviously kick off a major sync job.
I really have to say honestly: no shit. Sorry if i sound crass but really? When you use an arbitrary piece of software to automatically modify 100s of files the piece of software designed to detect that exact event has do do work and you are complaining? The mind boggles
If you are going to use git/Dropbox like that (and I can't say I really recommend it in the first place) then you should keep a bare repo in dropbox and use it as a local 'remote' (check out working repos from it on each computer that you are working on, and push/pull to/from the bare repo in Dropbox).
The only takeaway is that his power usage shouldn't be that high with Dropbox idling/paused. But given that he's
* surprised at it using a lot of power when doing the initial sync
* decided to have a git repo inside Dropbox (just don't do it, you will run into trouble; and it will cause heavy sync activity, thus waking CPU and wifi)
* using a unit of capacity to express power consumption
I'm not sure I trust his methods.
I regularly turn off Dropbox on and off when switching between mobile broadband and wifi, and haven't seen any difference in power consumption. That's on Ubuntu 13.04 @ Thinkpad T530.
Every modern operating system has inotify-like capabilities. Mac OS has an integrated desktop search since 2005, which wouldn't be feasible without filesystem events.
Dropbox has its own process though, doesn't it? dbfseventsd.
sudo fs_usage -w
I've been watching it recently and it seems to touch files that it shouldn't need to. I need to do more investigation. I uninstalled backblaze recently because I discovered that it manually scans the whole filesystem _all the time_.
Thanks for the Backblaze info! That's really good to know. Do you know if CrashPlan is any saner? Resource usage was a big factor when deciding between the two, and I want to know if I've made the right choice :)
Interesting question - I hadn't gotten around to checking the numbers yet myself. So,
Indexes are stored in S3, data is on Glacier
~450GB of data stored
316,189 Requests to upload it all (main cost)
$17.39 to upload it all
$5.09 to store it last month (but upload was incomplete)
Expect closer to $6-$7 to store it this month
~$120 to restore it all (that was a back of the napkin figure before I started)
Arq has a number of good things to recommend it. The creator lurks around here, which is always nice to know. The design is very sensible - it's basically a git index on S3 with the data blobs in Glacier. The storage format is open and documented. All the data is encrypted using your own key locally and then pushed to your own AWS account. It works out really cheap if you use Glacier (only planning on restoring in catastrophic circumstances).
It all just works. The interface is easy and intuitive. I pushed my data to the EU region since it's closer. The more I think about it the happier I am with my choice at the moment. Even if I changed my mind and stopped using it - it's nice knowing that for a few dollars a month I have a huge snapshot of my data in a reasonable format in my AWS account.
My method is not scientific at all, I agree with that BUT the facts remain: without dropbox on (even after full sync and supposedly idle), my battery life is multiplied by nearly 3.
As a side note, Dropbox behaves differently on OS X. I've never noticed this issue on Windows 8 machines.
One solution to the whole "storing source in dropbox" problem that I liked to use previously (using bitbucket now instead) is just to have a bare git repo in dropbox + the real checkout somewhere else and just push commits to the bare dropbox repo whenever I'm about to switch machine. Could even be set up to happen more automagically since the repo isn't shared.
Syncing binaries (I mainly work in C#) and lots of small changes seems silly to me.
I switched from Dropbox to SpiderOak recently, and they're a bit better, but feels so clunky. I'm just about fed up with all of them. I use CrashPlan for backups and they're at least good for CPU usage because they have a preference -- max CPU (%) when you're at the computer, and max CPU when you're away, which really does help.
Those who are using Dropbox may find something like cpulimit[1] useful. This limits the CPU usage of any process by repeatedly sending SIGSTOP and SIGCONT signals to your process. It's on homebrew for OS X. I think there're some other alternatives available too but can't remember off hand. But if the article's accurate, Dropbox has issues even when it's not indexing heavily.
A non-open-source OS X app called AppTamer can be very handy for managing battery life. You can set process priority and automatically pause non-foreground GUI apps. So when you switch away from your CPU chewing IDE, it'll get SIGSTOP'd.
"I just have to change my habit of having it constantly on while working. Turn Dropbox on and sync before starting to work and then turn it off."
Hohoho. Allow me to suggest a solution. Suppose that you would be ok with having Dropbox sync files every five minutes. Let's suppose that 10 seconds is usually more than enough to do whatever syncing is necessary. Well, then:
$ while :; do killall -STOP Dropbox; sleep 300; killall -CONT Dropbox; sleep 10; done
I do things like this to throttle the CPU usage and, in some cases, the power consumption of applications that really don't need what they're using. This really should be a feature provided by the operating system, but in the meantime, I resort to ridiculous hacks of this sort. (Note that, if you ^C the above loop, Dropbox will likely be still STOPped and you'll have to CONT it again. It is possible to write a more sophisticated program to address this problem.)
The Dropbox daemon wakes up every 10-15 seconds, uses 15% of the CPU then goes back to sleep. All this without any files changed or the index being affected. On 10.8, 10.9 with official versions and betas. Noticeable because I am on an older C2D MacBook Pro.
Wondering whether it's a side effect of something else installed e.g. iStat Menus.
This is interesting, I've also got a 2013 MBA and use DropBox but I've not seen a big dip in battery usage. I might do a bit more testing this week and see whether it has had an impact and I've just not realised, and/or I've put it down to the code-compile-simulator cycle.
I see this problem happening only to me when it's used as code repository. Which is what he is doing. I had that for one project because the client didn't want to use git, but it's horrible. So, logically, when you have a lot of files changing all the time (compiling (xcode, eclipse), adding libraries (node, RoR)) you'll see dropbox taking over your computer. For all other uses it seems fine.
Note also that when you have the issue with a lot of small files, during/after dropbox is done, Finder / mds will jump to the top of your top indexing the files. I disabled mds in my projects directory and will never place active source directories in dropbox again.
You are right. That is strange. I never tried turning synching off; I just turn Dropbox off entirely. But is that load not a result of the FS monitoring? Or does that take far less in itself?
Can anyone confirm his findings? I imagine that if this is real, then this is a problem for all MacBook owners not just the latest MacBook Air.
I have a 13" MacBook Air, medio 2011. I bought it almost right after it was released and have used it as my only computer almost every day since. I have the Dropbox agent running all the time.
While working last Sunday I got 5.5 hours of battery life out of it and I've never taken good care of the battery. I never drain it unless I'm not near a power outlet, in which case I just have it plugged in all day.
My current battery health information shows:
Cycle Count: 253,
Condition: Normal
I have a cron script running on my MacBook to make sure the Dropbox client priority is always set to 20 (lowest). I also pause the client manually when moving to battery power. I really wish there was a better solution, like switching to a similar service that offers decent (or at least acceptable) software, but almost everyone else uses the same setup with terrible 'universal' Java/Python/etc-based clients that absolutely suck at synchronizing large amounts of data.
OS X 10.9 does show battery use per app in the Activity Monitor. It also shows a short list of battery-draining apps right in the battery menu at the top of the screen.
Open Dropbox when you need to sync, then quit the process. Make sure it doesn't run at OS X start up. I don't see why most of the users want to have a perfect sync all day long since it doesn't happen in real time (large files, slow network, etc.)
Oh if only it was one App that killed the battery on the last laptop I bought. Before nuking the default windows install, I did a quick test, with all the crapware that comes as standard, it managed about 4 hours. Fresh install 6.5 hours.
Since I'm stuck using an old computer I made a small script that turns Dropbox on every few hours and kills it after 10 minutes. I don't sync data all the time so this solution works well for me and I save system resources.
I push to a bare repo in a dropbox folder because I can't for the life of me get a secure connection through the work firewalls / proxy so I don't bother and let dropbox deal with it.
Yeah, it sucks big time. DB is at the center of my workflow. Having my git repository in it, my projects plans, etc. I just have to not forget to "turn on, sync, turn off, work, turn on sync, turn off".
On the other hand, I didn't notice any difference in battery consumption on my Windows 8 laptops. Who knows, maybe OS X Mavericks will change that with the way it handles processes?
Last I tested, Dropbox consumed CPU for ANY changes ANYWHERE in your filesystem, not just within the Dropbox folder. So its (slow) code processed changes even when you untarred a large set of files somewhere in /tmp.
I also stopped using Dropbox for storing my code because of that — it's nearly useless, as it consumes so much CPU and eats so much battery life, that it simply isn't worth the effort.
Another problem with Dropbox is that upon login it indexes the entire Dropbox, which if you have an HDD basically kills the machine for several minutes, e.g. you can't do anything else. That happens to me on a Mac Pro. On an SSD-equipped MacBook things are better, but it's still a monstrous operation.
And to everyone that says you should not store code in dropbox: a) there are good and legitimate reasons for doing so, b) dropbox could work a little more on optimizing CPU usage instead of adding "save your photos to Dropbox" features.