January 24, 201115 yr Just a simple question; is the speed of my array (with a parity drive in place) limited by the speed of the slowest drive? For example, if I have an old SATA I drive mixed in with a bunch of faster modern SATA II drives, will the old drive limit the achieveable read and write speeds?
January 24, 201115 yr It will not limit the read speed of anything but itself. It will not limit the "write" speed of anything but itself. (unless it is the parity drive, involved in every "write") When writing to the array the speed is limited by the rotational speed of the slowest drive involved. 7200 RPM drives will have write speeds 30% faster than 5400 RPM drives.
January 24, 201115 yr First, SATA I vs. SATA II makes no difference ==> both interfaces are far faster than the data can stream off of the platters. But to answer your question ... (a) For reads, the read speed depends on the speed of the drive the data being read is on ... no other drives are involved in this. In most cases, the limiting factor for your data transfer is the network speed ... not the drive speed. (b) For writes, the speed is a function of the speed of both the drive being written to and the parity drive. To write one sector of data requires 4 disk operations: two reads and two writes. One each on the drive the data's being written to and the parity drive. So if, for example, one is a 5400rpm drive and one is a 7200rpm drive, the speed will be limited by the 5400 rpm drive ... although there are other factors (areal density of the drive; size of the drive's cache; etc.). If your network is a 100mb network, none of these matter -- both reads and writes will be limited by the network speed. If you're on a gigabit network, then you'll get the maximum possible write speeds, but your reads will generally be limited by the network speed (although the inner cylinders of 5400rpm drives will probably be slower than the network).
January 24, 201115 yr Author Just one other question on this while I think about it: I was planning on using an old WD Raptor 74GB as a cache drive. But when I ran preclear on it (and these number may mean nothing, I don't know) I got read and write of approx 62MB/sec at the start of each read/write, but on my WDEARS I got 112MB/sec. Do these numbers translate into the final write speeds, as they would suggest that even though the Raptor has a 10,000rpm spin speed, its slower than the more modern WDEARS which would make it a worse choice as a cache drive?
January 24, 201115 yr Do these numbers translate into the final write speeds, as they would suggest that even though the Raptor has a 10,000rpm spin speed, its slower than the more modern WDEARS which would make it a worse choice as a cache drive? Yes... it indicates the old Raptor may spin faster, but be slower overall. The new SATA drive is faster... at least on the outer cylinders where bit density is greater.
January 24, 201115 yr Have a read over here -- http://lime-technology.com/forum/index.php?topic=10352.0 The last script posted by WeeboTech [ http://lime-technology.com/forum/index.php?topic=10352.msg98444#msg98444 ] is what you want to use to get a better idea of performance.
January 24, 201115 yr r.e. your Raptor speed .... The areal density on the EARS drives is greater than that on a Raptor, so it provides a faster transfer rate for data streaming from the drive ... ... HOWEVER ==> the Raptor's average seek time is far better than an EARS ... ~ 4.5ms vs. ~ 12ms ==> so a Raptor will START a write well before an EARS will. Note a Raptor can write ~ 1/2 MB before the EARS will even start the write ... and can likely keep up with the network transfer speed of the incoming data -- so I doubt the EARS would ever catch up. If the drives were being accessed locally, the EARS would be better for large file transfers, since it would "catch up" and then pass the performance of the Raptor ... but with an UnRAID server that's limited by the network transfer speeds (even on a gigabit network you're unlikely to exceed about 70-80MB/s) the "head start" the Raptor has would not likely be overcome by an EARS. Bottom line: The Raptor is a fine choice for your cache drive.
January 24, 201115 yr Garycase, even then, that's not always the case. It depends on the size of your writes and how much memory you have available. In my case the initial writes to my parity protected array with EADS are 234 MB/s for the initial 1.2 GB. The Raptors being able to start writing 512K before the EADS can start means nothing in light of my 1.2 GB.
January 28, 201115 yr I agree (as I said earlier) that with very large writes the EARS drives would "catch up" and pass a Raptor ... but that's unlikely to be the case in most scenarios, due to the network speed limitations. Even a finely tuned GB network will never exceed ~ 90GB/s ... and even if it was "perfect" (with no packet overhead and no retransmissions) it could only hit 125GB/s (125GB x 8 bits/byte = 1,000gb ... or a "perfect" transfer rate). EADS drives won't come anywhere close to 234MB/s -- that's FAR faster (more than double) than the platter streaming speed ... and 1.2GB is FAR more data than the 32MB cache can hold, so there's simply no way you can write at 234MB/s. I've seen outer cylinder READS hit ~ 115MB/s on an EADS or EARS drive on an internal SATA controller ... but even these will be limited by the network speeds when read from an UnRAID server.
January 28, 201115 yr And yet that's the speed that my system encounters on local file writes. [ http://lime-technology.com/forum/index.php?topic=10352.msg98458#msg98458 ] You can thank the Linux drivers and their use of memory as drive cache. That was my point. It depends on the size of your writes and how much memory you have available.
January 28, 201115 yr Author Right, this is a little over my head guys, but if I get the gist of what you are saying, having a cache drive is largely pointless unless a) you have very old, slow data drives (slow enough with parity to drop below the normal network speed) combined with a very fast cache drive, b) very little system memory. Given that I have WD EARS as data and parity drives, and 4GB of system memory, I'm assuming from what you've said, my Raptor cache drive won't really add anything. The only thing I can think it would add, is if I want to add a file to the array, whilst it is being used for streaming (presumably writing to it stops it being able to stream)?
January 28, 201115 yr No, that's not at all true. A cache drive will definitely improve your performance IF you have a GB network. On a 100MB network both your read and write speeds will be network limited (to ~ 11.5MB/s) ... so there's no reason for a cache drive. On a GB network you'll see write speeds to the protected array in the 30MB/s range -- less than half of what you could achieve writing to a cache drive. Remember that a single sector write to the protected array requires 4 disk operations, whereas a write to the cache drive is a single operation. Reads will generally be network-limited, but can get close to the disk transfer rate ... I usually see between 70 & 90MB/s. If you had a cache drive your write speeds would be close to those read speeds -- depending on the performance of the cache drive. Your Raptor would work very well for this purpose. VERY simple to test -- just install the Raptor (but don't assign it to the array) ... then write a few GB to a share on the array and note the transfer rate (or just time it); then shut down the array and assign the Raptor as a cache drive; then start it back up and write a few GB to the share again The 234MB/s described by Brit is completely useless with regard to UnRAID performance. That's a local Linux speed test -- NOT a test of access to the server ... i.e. no network involved. The high initial speed is NOT writing to the disk ... it's simply the speed at which data can be buffered in RAM. Note that by the time the 10GB test is written, the average write speed is 34.8MB/s ... and when it's read back the average read speed is 80.5MB/s. These are the REAL speeds the system is operating at.
January 28, 201115 yr One other comment -- just noticed your last line, "... The only thing I can think it would add, is if I want to add a file to the array, whilst it is being used for streaming (presumably writing to it stops it being able to stream)?" No, writing to the array doesn't stop it from streaming. UnRAID is quite robust -- you can actually run a parity check, write to the array, and stream a couple movies all at once. Certainly exercises the drives a lot if you do that ... but it works I don't actually use the array when running a parity check -- but I have written new files to the array while simultaneously streaming six different movies.
January 28, 201115 yr Author I don't actually use the array when running a parity check -- but I have written new files to the array while simultaneously streaming six different movies. Wow - 6 movies at once! That is impressive - where they all HD? Was that with SMB or NFS? I have a Netgear NeoTV (after switching from an older TViX 6500a) which has a 10/100 ethernet port on it. I can't seem to get decent streaming speeds using SMB (50-70Mbps) but get a rock solid 90Mbps over NFS - I assume it must be something to do with how the Netgear communicates using SMB. I am currently transferring my 3.5TB of data from my DroboFS to my new unRAID (parity drive not yet in place) and I am getting 35-37MB/sec -this may be limited by the read speed of the Drobo (which is pretty slow as NAS boxes go - hence why I am switching to unRAID). Once everything is set-up I will give the cache drive a go, thanks for your advice.
January 28, 201115 yr One HD, the others all standard DVDs (the vast majority of my collection are standard DVDs) ... using SMB. There are just two of us -- so the need to stream six at once is not very likely ... but I have six different places you can watch TV/movies (4 PCs tied to HDTVs, 2 extra computers), so I've tested with all six streaming at once -- all TV (from a Beyond TV Server), all TV (from a Sage server), all movies (from the UnRAID server), and combinations of the above. All of them work fine. Fully wired GB network with all Cat-6 cable. Writing data without a parity drive assigned is exactly like writing with a cache drive -- there's no extra overhead like there would be with fault tolerance in place. So your write speeds definitely seem slow ... either because the data's not coming off the Drobo any faster than that, or you've got some issue with your network that's causing the transfer speeds to be slow.
February 6, 201115 yr Resurrecting this thread as I just built an unRaid server with 2 drives and the speed is horribly slow at 8M/s when I transfer data from my PC to the unRaid. Here're my configuration: Parity Drive: WD Green 2TB Drive1: WD Green 1.5TB MB: Asus P5QPro RAM: 4GB DDR2 Both unRaid Server and my PC are connected to a Dlink gigabit switch via Cat5E cable. The Switch is connected to my wireless router. My PC configuration is: i7-950 with 2 Intel SSD in Raid0. Am I missing anything here? IF copying 200GB of data to the unRaid Server takes the whole day, what's the point of having it? Please help. Thanks.
February 6, 201115 yr Resurrecting this thread as I just built an unRaid server with 2 drives and the speed is horribly slow at 8M/s when I transfer data from my PC to the unRaid. Here're my configuration: Parity Drive: WD Green 2TB Drive1: WD Green 1.5TB MB: Asus P5QPro RAM: 4GB DDR2 Both unRaid Server and my PC are connected to a Dlink gigabit switch via Cat5E cable. The Switch is connected to my wireless router. My PC configuration is: i7-950 with 2 Intel SSD in Raid0. Am I missing anything here? IF copying 200GB of data to the unRaid Server takes the whole day, what's the point of having it? Please help. Thanks. 1. post a syslog 2. what speed is being negotiated between your switch and your server? type ethtool eth0 to find out. 3. What speed network interface exists between your PC and the router. To see, Start task manager and go to the network tab, there you can see the speed of the network connection. Is the disk you are copying from in the PC, or is it an external USB drive? What specific Dlink model switch do you have? Are you still calculating the initial parity calculation? Joe L.
February 6, 201115 yr 1. post a syslog How do I do it? 2. what speed is being negotiated between your switch and your server? type ethtool eth0 to find out. 3. What speed network interface exists between your PC and the router. My router speed is a 10/100MB. There're a megabit switch between router and everything else on my network. To see, Start task manager and go to the network tab, there you can see the speed of the network connection. It's said 100Mbps at my PC, Is the disk you are copying from in the PC, or is it an external USB drive? Yes, the disk i'm copying from in the PC is an internal drive. What specific Dlink model switch do you have? D-link DGS 2205 Are you still calculating the initial parity calculation? No. It took the whole day to do that at the initial setup already.
February 6, 201115 yr 1. post a syslog How do I do it? 2. what speed is being negotiated between your switch and your server? type ethtool eth0 to find out. 3. What speed network interface exists between your PC and the router. My router speed is a 10/100MB. There're a megabit switch between router and everything else on my network. To see, Start task manager and go to the network tab, there you can see the speed of the network connection. It's said 100Mbps at my PC, Is the disk you are copying from in the PC, or is it an external USB drive? Yes, the disk i'm copying from in the PC is an internal drive. What specific Dlink model switch do you have? D-link DGS 2205 Are you still calculating the initial parity calculation? No. It took the whole day to do that at the initial setup already. If it said 100Mb/s at your PC, then you are probably getting close to its maximum. (Theoretical max is somewhere between 10Mb/s and 12Mb/s on a 100Mb/s network.) If you are getting 8Mb/s it is not unusual. Does your PC have a Gigabit network interface? If it does, look to your PC configuration, or your wiring, or your network switch as to why it is negotiating only a 100Mb/s connection. Perhaps you have it plugged into your 100Mb/s router and not the switch. Perhaps you have a bad network cable. Joe L.
February 6, 201115 yr I double-checked everything and found that my PC network was only at 100Mbps thought it's connected to a megabit switch. It turned out the network catble used was only a cat5. I switched it to a cat5e cable and now the network is showing 1Gbps on my PC from the task manger window. The transfer rate is now doubled at 17MByte/second. Is it that best I can get from unRaid? QN
February 7, 201115 yr Sacman, 17MBytes/s is still low even if you only have WD Green drives. The reason is in your motherboard - it uses Atheros ship for the LAN and this one does not work well in Unraid (at least for 4.7). Unraid 5.xxx will use a newer kernel with perhaps better support for them but if you do not want to wait you can buy from a cheap Intel (or Realtek) based NIC and use this instead - do not forget to disable the onboard LAN.
February 8, 201115 yr A little update: annoyed by the slow speed, I went and replace all network cables with cat6 to match all the wiring in my house. This help tremendously, I'm transferring files as I'm typing this at 29MB/sec. It's still slower than I expected but it's certainly better than 7MBps or even 19MB/s. I think bcbgboy13 is probably right about my motherboard so my best option is to wait for the next version of unRaid. Hopefully, it'll address this problem. My plan is to stream media content to my PS3 via PS3MediaServer by hopping through my PC in the middle. If I decided to buy a NIC card, is there a recommendation? Also another question for bcbgboy13, why do I need to disable the onboard LAN? Is it a requirement of unRaid? QN
February 9, 201115 yr 29MB/s seems pretty reasonable to me. I'm seeing similar speeds running a gigabit/CAT6 network with an Asus M4A785-M board which uses a Realtek 8112L LAN chipset. Drives are WD greens and directly connected to the board. I don't have a cache drive, so writes are limited by parity. Then again, maybe we're both doing something wrong
February 10, 201115 yr I'm replacing an Intel S4200e NAS (using RAID 5 on four disks) w/ a custom built unRAID box. Everything on the network is identical (Cat6 cabling, all going through the same gigabit switch, etc...) and I get about twice the speed with the S4200. The one difference is I'm using green drives in the unRAID and 7200RPM disks in the S4200. My next step is to add an SSD cache drive to my unRAID. Hopefully that'll remove any disk bottle neck.
Archived
This topic is now archived and is closed to further replies.