What is true GigaBit speeds? I thought I had it

Posted:
in General Discussion edited January 2014
I have a netgear gigabit router

I bought a WD World Book NAS that's also gigabit.



I have cat6 running to each device



Intel Mac Pro Tower > Cat 6 Cable >> Gigabit Router << Cat 6 Cable < WD World Book



I'm coping a huge load i'm at 105.5 GB of 820GB, says 25 hrs, but the speed seems about 35% capacity of gigabit.. i'm copying from my mac pro to my WD NAS..



I have no other apps running.. should it really be taking that long.



I did a google search, they're saying that Cat6 is 600Mbps not true 1,000???



Do I need Cat7 Cable to get true Gigabit transfer speeds?
«1

Comments

  • Reply 1 of 26
    hirohiro Posts: 2,663member
    Quote:
    Originally Posted by domerdel2 View Post


    I have a netgear gigabit router

    I bought a WD World Book NAS that's also gigabit.



    I have cat6 running to each device



    Intel Mac Pro Tower > Cat 6 Cable >> Gigabit Router << Cat 6 Cable < WD World Book



    I'm coping a huge load i'm at 105.5 GB of 820GB, says 25 hrs, but the speed seems about 35% capacity of gigabit.. i'm copying from my mac pro to my WD NAS..



    I have no other apps running.. should it really be taking that long.



    I did a google search, they're saying that Cat6 is 600Mbps not true 1,000???



    Do I need Cat7 Cable to get true Gigabit transfer speeds?



    For reality you aren't doing too bad. Better cables may help a bit, but there are a couple issues that reduce user measured throughput compared to max theoretical throughput.



    One is numeric base transitions 1Gbit is one billion bits (1 x 10^9), but when you transfer files and the OS displays the results the file size is measured in binary so the binary equivalent is (2^30). 2^30 == 1,073,741,824 which is bigger than 1,000,000,000, so even when everything is running at max theoretical you could only measure an apparent ~93% throughput.



    Second, you could only get max throughput if there are no collisions, collisions can bring a network entirely to all stop if you try to push somewhere in the neighborhood of 70% capacity through it, because every collision doubles the amount of bandwidth that packet takes.



    Third, unless you are doing something like using unreliable UDP, the receiving machine sends back an acknowledgment, a packet totally unmeasured by a naive throughput metric. Those ACKs also raise the collision probability significantly. Double trouble for the throughput measurement.



    There are other router related network issues, things that are overall very good for connecting to the world, but may help sub-optimize (just a little) a local machine to machine transfer, especially if you have multiple machines connected to the router. All that backbone overhead doesn't consume much bandwidth itself, but it does impact collision probability when you try to max out a local file transfer.



    So a true 350Mbit isn't horrible. I'm sure there are some sysadmins around who can provide specifics on how to improve that some but you won't ever see anything near true Gbit transfer speeds using TCP traffic on a 1Gbit system.
  • Reply 2 of 26
    Quote:
    Originally Posted by Hiro View Post


    For reality you aren't doing too bad. Better cables may help a bit, but there are a couple issues that reduce user measured throughput compared to max theoretical throughput.



    One is numeric base transitions 1Gbit is one billion bits (1 x 10^9), but when you transfer files and the OS displays the results the file size is measured in binary so the binary equivalent is (2^30). 2^30 == 1,073,741,824 which is bigger than 1,000,000,000, so even when everything is running at max theoretical you could only measure an apparent ~93% throughput.



    Second, you could only get max throughput if there are no collisions, collisions can bring a network entirely to all stop if you try to push somewhere in the neighborhood of 70% capacity through it, because every collision doubles the amount of bandwidth that packet takes.



    Third, unless you are doing something like using unreliable UDP, the receiving machine sends back an acknowledgment, a packet totally unmeasured by a naive throughput metric. Those ACKs also raise the collision probability significantly. Double trouble for the throughput measurement.



    There are other router related network issues, things that are overall very good for connecting to the world, but may help sub-optimize (just a little) a local machine to machine transfer, especially if you have multiple machines connected to the router. All that backbone overhead doesn't consume much bandwidth itself, but it does impact collision probability when you try to max out a local file transfer.



    So a true 350Mbit isn't horrible. I'm sure there are some sysadmins around who can provide specifics on how to improve that some but you won't ever see anything near true Gbit transfer speeds using TCP traffic on a 1Gbit system.





    WoW! first off thank you for the elaborate response. Here's the Get info off the mapped Volume Drive:

    afp://MyBookWorld._afpovertcp._tcp.local/Public



    so it's some translation of AFP to TCP ? How much is lost there?



    I'm also wondering if i just plugged it in directly to the secondary ethernet port on my mac pro.. for NAS it could be a straight patch or cross-over? i'm not saying I'll do that, but in the future, would it:



    1. less of collision for higher speeds

    2. need crossover?



    I'd try it right now, but i'm making a massive transfer as we speak.
  • Reply 3 of 26
    Unfortunately, Hiro's response is, um, mostly wrong.



    First, you don't have any collisions (well, unless there's something wrong with the Mac or the Router). I'm not going into long detail, but feel free to lookup collision domains. Collisions are _pretty much_ only an issue on shared networks (read: hub, not switch).



    Your issue is mostly hard drive read and write limitations. A fair average sustained read throughput would be in the 60-75 MB/s, and average write throughput is in a similar range. This assumes both drives are modern 7200 RPM drives. Both read and write speeds have increased a lot in the past couple of years, so even a "top of the line" drive that's 2 years old might be outperformed by a middle of the road drive released today.



    Then, you need to consider the overhead associated with packetization of data into TCP...the theoretical max data throughput (sometimes called "goodput") is .94% of the link speed. A much more realistic average is about .75% of link speed.



    So, you're looking at something like 60MB/s * .75 = 45MB/s = 360Mbps



    If either drive is a bit slow, that 360Mbps starts dropping pretty quickly. There's also a lot of variability in read and write speeds from the beginning to the end of a drive.



    Hiro is right in mentioning GB vs GiB...If the OS reports 105.5GB that's actually about 113,279,762,432 Bytes. Multiply that by 8 to get 906,238,099,456 bits.



    906,238,099,456 bits / 360,000,000bps = about 42 minutes in a perfect world



    I'd be really curious to hear how long your transfer actually took.







    Quote:
    Originally Posted by domerdel2 View Post


    WoW! first off thank you for the elaborate response. Here's the Get info off the mapped Volume Drive:

    afp://MyBookWorld._afpovertcp._tcp.local/Public



    so it's some translation of AFP to TCP ? How much is lost there?



    I'm also wondering if i just plugged it in directly to the secondary ethernet port on my mac pro.. for NAS it could be a straight patch or cross-over? i'm not saying I'll do that, but in the future, would it:



    1. less of collision for higher speeds

    2. need crossover?



    I'd try it right now, but i'm making a massive transfer as we speak.



  • Reply 4 of 26
    MarvinMarvin Posts: 14,208moderator
    Quote:
    Originally Posted by domerdel2 View Post


    Do I need Cat7 Cable to get true Gigabit transfer speeds?



    Faster hard drives. If your drives read/write at around 80MB/s, which is near the upper end of a 7200 rpm drive, your transfer rate can only ever be 80 x 8 = 640Mbps.



    If you get a RAID system, you might push it higher but there are some network overheads already mentioned.
  • Reply 5 of 26
    Quote:
    Originally Posted by concentricity View Post


    Unfortunately, Hiro's response is, um, mostly wrong.



    First, you don't have any collisions (well, unless there's something wrong with the Mac or the Router). I'm not going into long detail, but feel free to lookup collision domains. Collisions are _pretty much_ only an issue on shared networks (read: hub, not switch).



    Your issue is mostly hard drive read and write limitations. A fair average sustained read throughput would be in the 60-75 MB/s, and average write throughput is in a similar range. This assumes both drives are modern 7200 RPM drives. Both read and write speeds have increased a lot in the past couple of years, so even a "top of the line" drive that's 2 years old might be outperformed by a middle of the road drive released today.



    Then, you need to consider the overhead associated with packetization of data into TCP...the theoretical max data throughput (sometimes called "goodput") is .94% of the link speed. A much more realistic average is about .75% of link speed.



    So, you're looking at something like 60MB/s * .75 = 45MB/s = 360Mbps



    If either drive is a bit slow, that 360Mbps starts dropping pretty quickly. There's also a lot of variability in read and write speeds from the beginning to the end of a drive.



    Hiro is right in mentioning GB vs GiB...If the OS reports 105.5GB that's actually about 113,279,762,432 Bytes. Multiply that by 8 to get 906,238,099,456 bits.



    906,238,099,456 bits / 360,000,000bps = about 42 minutes in a perfect world



    I'd be really curious to hear how long your transfer actually took.



    Thank you all for the input. Keep subscribed to this post and i'll let you know the amount and duration it took to transfer as is.
  • Reply 6 of 26
    Quote:
    Originally Posted by Marvin View Post


    Faster hard drives. If your drives read/write at around 80MB/s, which is near the upper end of a 7200 rpm drive, your transfer rate can only ever be 80 x 8 = 640Mbps.



    If you get a RAID system, you might push it higher but there are some network overheads already mentioned.



    So if the Drive is SATA II, that speed is rated as transfer READ? the actual WRITE speed is based on the rpms? (which it is a 7200 rpm drive)
  • Reply 7 of 26
    MarvinMarvin Posts: 14,208moderator
    Quote:
    Originally Posted by domerdel2 View Post


    So if the Drive is SATA II, that speed is rated as transfer READ? the actual WRITE speed is based on the rpms? (which it is a 7200 rpm drive)



    No, both read and write speeds are limited by the physical storage process - spindle speed, areal density and so on - read speeds are usually faster than write but are typically within 0-100% faster than the write speed. The SATA number is for the interface between the drives, which naturally should be in excess of what the drive is capable of to allow the drive technology to advance.
  • Reply 8 of 26
    hirohiro Posts: 2,663member
    Quote:
    Originally Posted by concentricity View Post


    Unfortunately, Hiro's response is, um, mostly wrong.



    First, you don't have any collisions (well, unless there's something wrong with the Mac or the Router). I'm not going into long detail, but feel free to lookup collision domains. Collisions are _pretty much_ only an issue on shared networks (read: hub, not switch).



    Your issue is mostly hard drive read and write limitations. A fair average sustained read throughput would be in the 60-75 MB/s, and average write throughput is in a similar range. This assumes both drives are modern 7200 RPM drives. Both read and write speeds have increased a lot in the past couple of years, so even a "top of the line" drive that's 2 years old might be outperformed by a middle of the road drive released today.



    Then, you need to consider the overhead associated with packetization of data into TCP...the theoretical max data throughput (sometimes called "goodput") is .94% of the link speed. A much more realistic average is about .75% of link speed.



    So, you're looking at something like 60MB/s * .75 = 45MB/s = 360Mbps



    If either drive is a bit slow, that 360Mbps starts dropping pretty quickly. There's also a lot of variability in read and write speeds from the beginning to the end of a drive.



    Hiro is right in mentioning GB vs GiB...If the OS reports 105.5GB that's actually about 113,279,762,432 Bytes. Multiply that by 8 to get 906,238,099,456 bits.



    906,238,099,456 bits / 360,000,000bps = about 42 minutes in a perfect world



    I'd be really curious to hear how long your transfer actually took.



    Collisions are ALWAYS a possibility when using any reliable protocol, the ACK transits over the same wire as the packets. It's not a problem at all at low utilization, but IS one of the main reasons it is mathematically impossible to fully utilize the link at it's rated capacity. OP also said he was using a router, not a switch, which pretty much invalidates the naive hope you display of no collisions at all based on a switches backplane speed. The router actually raises the collision possibility by an tiny incremental amount, simple math of adding another transmitting-receiving microprocessor on the network.



    Your packetization rate is backwards too. The network throughput is not limited to less than or equal to the drive bus speed, the drive production rate is just the max date rate input into the network controller where the packetization overhead is added on top of the drive throughput for the theoretical max transmission rate outbound from the computer. You also forgot to translate from MB/s to Mb/s making your numbers VERY inaccurate. So the real rate in your example is 60MB/s / .75 == 80MB/s which translates to 640Mb/s. Your incorrect figure was seemingly almost a match for the OP's reported throughput, but here we see the real figure for your calculation identifies a missing 30% of capacity. Where'd it go?



    I'm not wrong, ACK collisions and even ACK explosions are a known issue, the higher the utilization of a pipe, the more contention for that pipe. Its a mathematical certainty. No opinion will ever change that. Especially when the opinion was built upon utterly incorrect mathematics. But alas, on further review this isn't the probable root cause.



    Looking up the WD Book specs (which would have been good to do before), it is a pathetically slow drive for NAS, which is where I agree real issue is. Don't fall into the trap of believing the laws of physics and protocols for wire transfer don't apply for ALL sizes of networks though, and be real careful with the math whenever you try to use it as an example.
  • Reply 9 of 26
    Quote:
    Originally Posted by Hiro View Post


    Looking up the WD Book specs (which would have been good to do before), it is a pathetically slow drive for NAS, which is where I agree real issue is. Don't fall into the trap of believing the laws of physics and protocols for wire transfer don't apply for ALL sizes of networks though, and be real careful with the math whenever you try to use it as an example.



    Umm.. I did, and not sure how I developed this was "pathetically" slow (a bit subjective depending on the user)



    http://www.amazon.com/Western-Digita.../ref=de_a_smtd



    Some highlights, of the specs:

    Quote:

    Cooler, quieter, eco-friendlier - Equipped with WD drives using WD GreenPower Technology, this system consumes significantly less power and delivers performance on par with power hungry network storage devices using 7200 RPM hard drives.



    Capacity gauge - See at a glance how much space is available on your drive.



    High-speed data transfer with Gigabit Ethernet - Gigabit networking and transfer rates are five to six times faster than other standard 10/100 Ethernet network storage systems. This performance is comparable to USB 2.0 direct-attached storage.



    Technical Details

    Form Factor: Desktop

    RJ-45 Ports: 1

    Interface Type: (10/100/1000) Gigabit Ethernet

    Data Transfer Rate: 10/100/1000 Mb/s

    Protocols: HTTP, HTTPS, CIFS/SMB, NFS, FTP, AFP




    Not saying this is the worlds fastest NAS by any means, but for 184 bucks out the door. I don't think the dollar to speed ratio is that pathetic. For an enterprise level of demand, I would agree with "pathetic".. for home network... not so much
  • Reply 10 of 26
    akacakac Posts: 510member
    Quote:
    Originally Posted by Hiro View Post


    OP also said he was using a router, not a switch



    Um, most routers have switches built-in. And the routing only occurs when routed from local to non-local networks. I am assuming he is transferring data on the same local network so it basically would just be the same as a switch.
  • Reply 11 of 26
    Quote:
    Originally Posted by Akac View Post


    Um, most routers have switches built-in. And the routing only occurs when routed from local to non-local networks. I am assuming he is transferring data on the same local network so it basically would just be the same as a switch.



    yup, Local network transfer
  • Reply 12 of 26
    hirohiro Posts: 2,663member
    Quote:
    Originally Posted by Akac View Post


    Um, most routers have switches built-in. And the routing only occurs when routed from local to non-local networks. I am assuming he is transferring data on the same local network so it basically would just be the same as a switch.



    Not consumer routers. You can buy switched ones, but not the most popular low priced ones, even with Netgear. Given a WD NAS and questions about throughput, do you think the OP is the knowledgeable discerning buyer that would be sure to buy a non-hub router? That's not criticizing domerdel2, it's more a criticism of the router manufacturers that they still sell hubbed routers for overinflated prices, without making sure everyone knows the differences, All in the name of maintaining product differentiation and wringing unwarranted extra $$ out of switched routers.
  • Reply 13 of 26
    Total file transfer 1,518 Gigabytes (1.48 TB)

    Time: 45 hrs 23 mins




    my brain is fried on the math of this.
  • Reply 14 of 26
    You're limited by the speed of the drives involved.



    Fast RAID arrays would help. They'd also cost a lot and require actual computer skills to set up.
  • Reply 15 of 26
    hirohiro Posts: 2,663member
    Just hook em up and walk away, that's all you can do as well as the general point of NAS in the first place. You get multi machine flexibility for the price of lower performance than a dedicated machine-bus connection.



    Thats about a 79Mbit/sec estimate. That includes a lot of overhead for small file (less than 1MB) accesses. If you have lots of larger files the actual transfer will be A LOT quicker.
  • Reply 16 of 26
    dfilerdfiler Posts: 3,420member
    This thread was an interesting read.



    However, it seems that much of the techno-mumbo-jumbo is poorly enough applied that I would advise novices to ignore it entirely. While we could go around in circles about collisions and wire-speed switching and hubs vs switches in consumer grade routers, that would likely be a disservice to anyone who might actually be trying to read this thread for an answer on the original topic.



    First off, cabling matters little in most scenarios. Unless your router tells you that packets are being dropped, Cat5e or even Cat5 is fine. There will be zero difference in performance. Long distance runs and runs next to electrical wiring is normally what necessitates the bump to Cat6. With that said, I typically use Cat6 for in-wall wiring because it would be expensive to replace once installed. For patch cables, i'll use whatever is handy.



    Also, I would forget the discussion of collisions. It is highly unlikely that your gigabit router has an internal hub instead of a switch serving the LAN ports.



    In the real world, with expensive hardware, you can acheive around 45MBps throughput. For consumer grade equipment, anything above 30MBps would make me happy.



    In consumer gear, it is typically the stuff connected to the network, rather than the network itself, that limits through put. One exception can be jumbo frame support. Believe it or not, the new i7 based iMacs (and perhaps other Macs) don't have jumbo frame support! Boy was I pissed when I found that out. My new quad-core Mac can never perform as fast as my previous two generations of Macs. The reason to not mix MTU sizes is that some devices drop drastically in throughput if needing to convert between sizes. This is true for my ReadyNAS NV+. It performed much worse until I set it back to 1500 MTU. In other words, it is worth confirming that everything on your LAN supports jumbo frames. Otherwise, my advice is to disable jumbo frames on everything.



    But really, i'd be looking at drive speed as the culprit. Or perhaps drive controller. Many cheap NASs can't even saturate a gigabit connection.



    Finally, unless dealing with a RAID, you will likely never saturate a gigabit connection except for brief moments. The reason is that the _write_ speed of drives, especially when dealing with numerous small files, will be the bottleneck. The write speed of the device you're writing to is almost always the limiting factor.
  • Reply 17 of 26
    I'm sorry Hiro, but this is just completely wrong. All routers are switches. Each and every port on even the cheapest home router is its very own network segment, with a collision domain limited to the device (computer, NAS, whatever) and the router. And even the most basic home router will do full-duplex, meaning that both the router and the device can transmit at full speed (in opposite directions of course) and NEVER have a collision.





    Quote:
    Originally Posted by Hiro View Post


    Not consumer routers. You can buy switched ones, but not the most popular low priced ones, even with Netgear. Given a WD NAS and questions about throughput, do you think the OP is the knowledgeable discerning buyer that would be sure to buy a non-hub router? That's not criticizing domerdel2, it's more a criticism of the router manufacturers that they still sell hubbed routers for overinflated prices, without making sure everyone knows the differences, All in the name of maintaining product differentiation and wringing unwarranted extra $$ out of switched routers.



  • Reply 18 of 26
    hirohiro Posts: 2,663member
    Quote:
    Originally Posted by concentricity View Post


    I'm sorry Hiro, but this is just completely wrong. All routers are switches. Each and every port on even the cheapest home router is its very own network segment, with a collision domain limited to the device (computer, NAS, whatever) and the router. And even the most basic home router will do full-duplex, meaning that both the router and the device can transmit at full speed (in opposite directions of course) and NEVER have a collision.



    Believe what you wish.
  • Reply 19 of 26
    dfilerdfiler Posts: 3,420member
    I've never seen a gigabit router based upon a hub rather than a switch. Is such a thing actually on the market?
  • Reply 20 of 26
    The very nature of a hub is that it acts as a repeating bridge (which is a clearly defined network device, not a general term here) between separate physical network segments, essentially making multiple wire segments act as one. There is _no_ routing going on or even possible in a hub. A hub literally just repeats all traffic on each segment to every other segment. It's about as fundamentally opposite as you can get to a router.



    Quote:
    Originally Posted by Hiro View Post


    Believe what you wish.



    It's not a belief, it's simple fact that anyone can look up. I really am sorry, I hope you don't take it personally, but factually wrong info isn't going to help anyone reading this thread.
Sign In or Register to comment.