Digicube - Page 4
Register

We are the best invite forum on the internet! Here you will find free invites, free seedboxes, free bonuses, and much more. Our members know the true meaning of sharing and have created a truly global bittorent community! Our site has the most up to date information on all private trackers and our members will guide you and introduce you to this truly secretive and enlightened club. Ready to get started? Register now!


Page 4 of 4 FirstFirst 1 2 3 4
Results 31 to 40 of 40

Thread: Digicube

  1. #31

    Join Date
    Mar 2010
    Posts
    87

    Default

    The cpu is only fast as the hard disk is, hashing requires the recheck of the data on the disk. Sometimes when your box hashes check its iowait you will note that the CPU waits for the disk to respond.



  2. To remove ads become VIP. Inquire about advertising here.
  3. #32

    Join Date
    Oct 2009
    Location
    UK
    Posts
    561

    Default

    You seem very keen on saying HDs are at fault as the HD is faster than the port I dont think thatll be a problem in uploading and in hashing it may be either.

  4. #33

    Join Date
    Feb 2010
    Posts
    44

    Default

    Long story short, the bottleneck usually isn't the hard drive, it's the processor.

    Basically, a torrent isn't pure I/O. All that overhead into piece checking, hashing, peer maintenance, etc. are all done by the processor. If your NIC is a cheap POS, your CPU also handles all TCP processing. For a processor targeted at consumers, the odds of getting a good NIC are slim. In case you're wondering, Atom is about 80% as fast as a Pentium 3 clock for clock. For example: a dual-core N330 is about as good as two Tualatin 1.3 GHz.
    Unless you're somewhere around a couple torrents a hundred GB or more with less than 100 peers each with 10/100 or better, your Atom is going to be your bottleneck before your hard drive.

  5. #34

    Join Date
    Feb 2010
    Location
    Finland
    Posts
    361

    Default

    Torrents are mainly bound to HDD I/O, IOPS particularly. Transfer rates are plenty, but seek times aren't. That's why you need as many HDDs as possible, even if they are smaller, with a large block size, so that as much as possible different HDDs can be doing different tasks, maximizing IOPS.

    CPU is mostly waiting for I/O in all of our boxes.

  6. #35

    Join Date
    Mar 2010
    Posts
    12

    Default

    Quote Originally Posted by verily View Post
    Long story short, the bottleneck usually isn't the hard drive, it's the processor.

    Basically, a torrent isn't pure I/O. All that overhead into piece checking, hashing, peer maintenance, etc. are all done by the processor. If your NIC is a cheap POS, your CPU also handles all TCP processing. For a processor targeted at consumers, the odds of getting a good NIC are slim. In case you're wondering, Atom is about 80% as fast as a Pentium 3 clock for clock. For example: a dual-core N330 is about as good as two Tualatin 1.3 GHz.
    Unless you're somewhere around a couple torrents a hundred GB or more with less than 100 peers each with 10/100 or better, your Atom is going to be your bottleneck before your hard drive.
    That is total rubbish. The bottleneck is the HDD IO not the CPU. When top/htop shows WA that means it is waiting for the disk to read, not the cpu or anything else. Even the most basic shitty OVH server has a good enough CPU to handle torrents (the exception is rar/zip which DOES use CPU).

    The Atom that Digicube offers is roughly as good as the pentium 4 2.4Ghz, and in fact uses the same architecture on a smaller production model.

    An Atom is only limited by RAM and HDD, and will do 100Mbit all day every day.

  7. #36

    Join Date
    Feb 2010
    Posts
    44

    Default

    Quote Originally Posted by Malachi View Post
    The Atom that Digicube offers is roughly as good as the pentium 4 2.4Ghz, and in fact uses the same architecture on a smaller production model.
    I disagree. By the way, if they were both the same, then that 1.6 GHz Atom couldn't possibly be equal to a Pentium 4 at 2.4 GHz.
    First off, Netburst was designed from the ground up to push clock speeds. It didn't double the P6 pipeline depth for no good reason. It didn't skip DDR and go straight to QDR FSB for the hell of it. It was originally designed to push 10+ GHz before the next generation. Its was designed under the old rule of 1% die space for 1%+ performance. It's a chip designed for high bandwidth to overcome its deficits.
    The Atom was designed from the ground up for low power. It has half the depth of the last Pentium 4 design, a completely different micro-op core with more complex instructions, less than half the ALU's, no integer mult/divide, half the issue width, no oooe, completely different L1 cache, the list goes on. Intel's design team used the rule of 1% power use for 1%+ performance, hence the frankenbeast of a design. Suffice to say the more notable traits the Atom share with Pentium 4 is limited to the FSB specs, L2 cache specs, and hyperthreading. Everything else shares more with Pentium (P5) than Netburst.

    In comparison, an Atom at 1.6 GHz is about 30-40% slower than a Pentium 4 at 2.4Ghz.
    Granted, it does so using about 1/5th the power.
    Note that means its performance is about as good as an architecture 8 years older.

    Like I said earlier, an Atom won't be your bottleneck with light torrenting. However, it will start choking under heavy use.
    Last edited by verily; April 7th, 2010 at 12:38 PM.

  8. #37

    Join Date
    Oct 2009
    Location
    UK
    Posts
    561

  9. #38

    Join Date
    Feb 2010
    Location
    Finland
    Posts
    361

    Default

    There's lots of different type of operations, MP3 encoding is one of them stressing certain components.

    ATOM by no means is fast CPU, but it's plenty fast for torrenting. People are happily torrenting even with embedded devices 1/5th of the CPU power of an ATOM. Even with 1.2Ghz ARM cpu people have reported being able to easily max out 100Mbps and staying mostly I/O bound.

    Torrenting itself is very lightweight on the CPU, it moves data, and when you move data you are I/O bound inherently. No amount of parallel processing (amount of torrent instances) is going to change that, no amount of arguing is going to change that. But if you change your old spinners to SSD then the scenario might be different, then again it will most likely be network I/O bound, remove that and then it might become CPU bound. But whoever has money for 10GigE & bunch of SSDs definitely can afford pretty much any CPU and then it's again I/O bound.

    Nothing is going to change the fact that torrent is dealing with data primarily and inherently thus I/O bound. Choosing peers, making network connections, reading or writing data, with properly designed application is so lightweight that the computations simply does not matter.

    The only time torrenting might be CPU bound is with slow CPU, and nothing else running on the box and you are hashing some large torrent so you get to maximize reading speed. Not sure about ATOM, but atleast VIA has AES in hardware, and if this is the case for ATOM too, the actual hash calculations are freakishly fast, and left I/O bound.

    It's true ATOM is slow, low power CPU, but much slower CPUs are handling torrents just fine as well.

    Superb Seedboxes starting from 5.06€ a month for TI members!


    Pulsed Media Rep - For Pulsed Media support contact via our site or via e-mail.

  10. #39

    Join Date
    Feb 2010
    Posts
    44

    Default

    To note, Atom doesn't have hardware encryption.
    Torrenting is a significant I/O activity. That's not at question here.
    The problem is I/O is still somewhat CPU dependent with respect to Atom boxes. No matter how good DMA is, it still doesn't completely eliminate the CPU. I seriously doubt that's a SAS drive in the specs (even then, you could measure usage if you tried). Granted, SATA has far less overhead than the ancient PIO modes, but the overhead is still measurable. Depending on the size of your torrent, it may not even be an issue since clients generally cache into main memory and the Digicube box is using 10/100, so the max I/O of ~10MB/s is far less than the average SATA disk's sustained I/O of around 20MB/s when going both ways.
    There is virtually no hardware offloading on chipsets that get paired with Atom which means network performance is also measurably CPU-dependent. At the very least, the base configuration doesn't use RAID. There is an option for it which I would strongly discourage because I seriously doubt they'll put a raid coprocessor into a cheap Atom box.

    I have no doubt people use weak CPU's for torrents and max out Fast Ethernet. Light torrenting isn't going to be bottlenecked by much more than network performance. It's when you start loading down with multiple torrents on a regular basis that the CPU becomes a problem.
    If you're the type to load a torrent once a week or so and seed it for a month or more, low end boxes built around Atom or worse are not going to be an issue (unless you're using something ancient like a Pentium MMX). If you're the type who keeps reloading torrents (even with FTP, client still has to hash it after you upload) the Atom is going to slow your box up to hours at a time for large torrents. If you're doing that while running several active torrents with hundreds of peers each, the Atom is going to choke hard.
    The only situations where a dedicated seedbox will be I/O bound by the hard drive involve gigabit or some really old clunkers that are more useful as magents. Any recent drive can handle multiple I/O and fully saturate Fast Ethernet as long as the CPU isn't tied down. However, pair an Atom onto gigabit and start running a few hundred torrents, you're just as likely to end up CPU bound.
    Last edited by verily; April 8th, 2010 at 12:51 PM.

  11. #40

    Join Date
    Oct 2009
    Location
    UK
    Posts
    561

    Default

    After running some tests on my home computer and the server checking the hash of a torrent at home uses about 5% cpu and 100% (100MB/s) I/O, on the server it uses 100% of both cpu and I/O so they are both bottlenecking it.

Page 4 of 4 FirstFirst 1 2 3 4

Similar Threads

  1. Digicube?
    By bluboy in forum Seedbox Discussions
    Replies: 4
    Last Post: September 29th, 2010, 11:34 AM

Tags for this Thread

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •