Would unRAID have any problems using the 4 port 1 Gb/s (and aggregating these ports) for 4 Gb/s or using a 10 Gb/s NIC?  Thanks!

unRAID supports a single ethernet interface. You are on your own for anything else.

 

Remember unRAID sustained write performance is not likely to approach a single 1Gbe interface.

Remember unRAID sustained write performance is not likely to approach a single 1Gbe interface.

 

I'm not convinced that your statement is correct.  I've seen in excess of 100MB/s on a 15GB file, writing to a cache drive.

 

In any case, a filestore appliance would normally spend more time serving files (ie. being read from) than being written to.  There is no reason why read performance should be any slower than writes to a cache drive.

I'm not convinced that your statement is correct.  I've seen in excess of 100MB/s on a 15GB file, writing to a cache drive.

 

In any case, a filestore appliance would normally spend more time serving files (ie. being read from) than being written to.  There is no reason why read performance should be any slower than writes to a cache drive.

 

Writes to cache are not sustained rates. If you which wish to see the sustained rate use a test exceeding cache size by a factor of 2.

 

Read performance should exceed write performance.

 

Remember unRAID sustained write performance is not likely to approach a single 1Gbe interface.

 

I'm not convinced that your statement is correct.  I've seen in excess of 100MB/s on a 15GB file, writing to a cache drive.

 

In any case, a filestore appliance would normally spend more time serving files (ie. being read from) than being written to.  There is no reason why read performance should be any slower than writes to a cache drive.

There are many factors involved. but i feel both statements do have truth in them... in Most cases, the average unraid server will not break the 1GB barrier or just bearly... even reading 2 streams at once.

 

But, there are those beast servers out there built on 7200 drives with SSD cache drives that can break the 1GB barrier. At least in theory...

I know those new seagates with the 1TB platters can read at about 180MB/s. better then a single GB line.  the question is can unRAID keep up with that?

 

I have no clue if a 10Gbe adapter is supported out of the box. I would think not, but you never know. I would just bind a set of 1GB nics and go from there. see if you saturate that. support for this was added to one of the recent beta's.

 

(If unraid does not support 10GB nics... ESXi does support 10GB nics and unraid in ESXi does support the 10GB virtual nic.. if you see where I am going with this...)

 

 

I'm not convinced that your statement is correct.  I've seen in excess of 100MB/s on a 15GB file, writing to a cache drive.

 

In any case, a filestore appliance would normally spend more time serving files (ie. being read from) than being written to.  There is no reason why read performance should be any slower than writes to a cache drive.

 

Writes to cache are not sustained rates. If you which wish to see the sustained rate use a test exceeding cache size by a factor of 2.

 

Read performance should exceed write performance.

 

When writing to an SSD cache, I do fully saturate my 1Gb line until it is 100% full, then it drops down to about 48MB/s. (Maybe i need a bigger cache )

 

To be honest.. unRAID was designed to be the opposite of a high performance storage system. super high read/writes were just not in its blood.

Writes to cache are not sustained rates.

 

I know what you are saying but, with a 500GB cache drive, exceeding the cache size (at least in my use of unRAID) is extremely unlikely.

 

If you which wish to see the sustained rate use a test exceeding cache size by a factor of 2.

 

Hmmm ... 3+ hours worth of transfer?

 

Read performance should exceed write performance.

 

Precisely!

I am using a 10Gb vmxnet3 adapter in my unRAID ESXi VM.  Quad port PT/1000 Intel adapter in the ESXi server set up in etherchannel mode (ip hash), dual port in my main desktop PC, also in etherchannel.  Switch ports set to etherchannel accordingly.  Transfer rates to a SSD cache drive peak at about 180-190 and level around the 153 mark.

 

So I guess unRAID is capable in theory - but obviously never directly to the array.  Screenshot of a file copy that I posted elsewhere:

 

 

 

 

 

 

 

When writing to an SSD cache, I do fully saturate my 1Gb line until it is 100% full, then it drops down to about 48MB/s. (Maybe i need a bigger cache )

So, the sustained write is well under 1Gbe.

 

but obviously never directly to the array.

 

Seems not so obvious, as you keep trying to hide the fact that unRAID will not sustain writes near 1Gbe.

 

First you say it wont, then you post a write to cache image. Maybe you should rethink.

 

IP hash would prevent a single stream from using multiple interfaces.

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1004048

Note: One IP to one IP connections over multiple NIC is not supported. (Host A one connection session to Host B uses only one NIC).

 

I realize most of you don't do large scale storage or networking, but don't make stuff up.

To be honest.. unRAID was designed to be the opposite of a high performance storage system. super high read/writes were just not in its blood.

 

Bingo!

 

I use unRAID. I like it. But I would not invest in 10Gbe infrastructure for an unRAID platform (yet$). Nor would I use link aggregation for a home network, the stream count/address space is too small. I have tried. However there is a 10Gbe switch (without buffering) due soon at a very reasonable price, Romley.

Thanks for all the input guys!

 

I should add the reason I initially asked about a higher speed interface was because I needed to upload at least 10 TB of data to Crashplan.  Doing this on a gigabit connection would take months, so I was looking for a faster solution.  This wouldn't be a permanent solution, and would just be used for the initial upload.  At this point though, it doesn't look as there are any easy out of the box solutions?

Thanks for all the input guys!

 

I should add the reason I initially asked about a higher speed interface was because I needed to upload at least 10 TB of data to Crashplan.  Doing this on a gigabit connection would take months, so I was looking for a faster solution.  This wouldn't be a permanent solution, and would just be used for the initial upload.  At this point though, it doesn't look as there are any easy out of the box solutions?

 

Sure buy a 10TB cache drive since

 

I know what you are saying but, with a 500GB cache drive, exceeding the cache size (at least in my use of unRAID) is extremely unlikely.

 

Right Peter?

 

 

OK, since you want to load 10TB, disable parity and you can get better performance. You are still looking at several months.

 

out of the box, you can access each drive separately and fill them (reiserFS). What is the source?

OK, since you want to load 10TB, disable parity and you can get better performance. You are still looking at several months.

 

Several months? Really?

Seems not so obvious, as you keep trying to hide the fact that unRAID will not sustain writes near 1Gbe.  First you say it wont, then you post a write to cache image. Maybe you should rethink.

 

What's to rethink?  With parity calculations going on, writing directly to the actual array is never going to be that quick.  That's all I was saying.  The cache drive isn't really a member of the array to my way of thinking, as it isn't involved with parity, just a dumb disk to write to.  But please feel free to educate me further, I welcome new learning experiences.

 

I realize most of you don't do large scale storage or networking, but don't make stuff up.

 

...and how about you be a little less abrasive and full of yourself?  I never claimed to know 100% what I'm talking about, I'm simply presenting what I have set up and what my system claims to achieve in terms of speed etc.  Note that I even said 'I GUESS it's capable, in theory'.  That's what people do when they see something happen, and not having the full detail, make an assumption based on what they have seen.

 

Thank you for presenting the facts above - it will certainly make me do some more research in an effort to understand this better.  Please work on your delivery though.

Seems not so obvious, as you keep trying to hide the fact that unRAID will not sustain writes near 1Gbe.  First you say it wont, then you post a write to cache image. Maybe you should rethink.

 

What's to rethink?  With parity calculations going on, writing directly to the actual array is never going to be that quick.  That's all I was saying.  The cache drive isn't really a member of the array to my way of thinking, as it isn't involved with parity, just a dumb disk to write to.  But please feel free to educate me further, I welcome new learning experiences.

 

I realize most of you don't do large scale storage or networking, but don't make stuff up.

 

...and how about you be a little less abrasive and full of yourself?  I never claimed to know 100% what I'm talking about, I'm simply presenting what I have set up and what my system claims to achieve in terms of speed etc.  Note that I even said 'I GUESS it's capable, in theory'.  That's what people do when they see something happen, and not having the full detail, make an assumption based on what they have seen.

 

Thank you for presenting the facts above - it will certainly make me do some more research in an effort to understand this better.  Please work on your delivery though.

 

I simply stated that unRAID sustained write performance is not likely to approach a single 1Gbe interface. You decided you knew better and put up some yet to be explained numbers about a completely different situation. You don't like facts, don't blame me.

 

1) Explain how you can claim 180MB/sec. You have described the equipment and software. Don't the words "IP hash" give you a hint?

 

2) What theory are you guessing? unRAID is capable of ? Are you guessing a theory for RAID calculations based on your example of cache drive performance? The performance of COW vs XOR on striped and non-striped datasets? Is it really the calculation that determines unRAID performance or is disk access overhead a bigger factor?

 

I think you are full of yourself. I provided information and documentation against your claims.

You have missed the point entirely - read my original post, I wasn't arguing with you at all, I simply presented what I had set up and what my systems claim in terms of transfer speeds.  Next you're going to claim I doctored that screenshot or something.

 

I DO like facts, in fact I have already stated that I am more than willing to be educated further and welcome the opportunity.  My "GUESS" (and please do take that definition literally) was based on what I saw occur, nothing more.  I wasn't theorising anything, as I don't have the appropriate level of technical expertise to present a theory.  Perhaps Teracopy (the application) doesn't actually report correct transfer speeds?  I don't know. 

 

I fully admit I am no storage/networking guru, but seriously mate, you have some sort of god complex.  How about engaging in a civil debate instead of being a complete jerk?  We could have discussed this like adults and worked out what exactly was going on in order for me to see the screenshotted transfer speeds, yet instead you choose to speak to me like one would a dog.  (Actually I wouldn't even speak to my dog like that.)  I have seen you lord it over many people on this forum in many different posts and it's pretty bloody ordinary to say the least.

 

It's a real shame as I truly do like learning - but not from someone who hoards his knowledge and only drip feeds them bits and pieces to make himself feel superior.  A sad case indeed.  Feel free to ignore my post btw, I will be doing you the same courtesy for any further posts from you.

OK, since you want to load 10TB, disable parity and you can get better performance. You are still looking at several months.

 

Several months? Really?

 

The OP's estimate was months.

Doing this on a gigabit connection would take months, so I was looking for a faster solution.

So, even with parity disabled, yes the estimate is probably still valid. My experience with Crashplan does not allow me to change that estimate.

 

On the other hand, "out of the box", if the source(s) can write to the reiserFS directly (SATA), perhaps in parallel, the time required will be reduced.

You have missed the point entirely - read my original post, I wasn't arguing with you at all, I simply presented what I had set up and what my systems claim in terms of transfer speeds.  Next you're going to claim I doctored that screenshot or something.

You posted contradictory information. That is arguing.

I DO like facts, in fact I have already stated that I am more than willing to be educated further and welcome the opportunity.  My "GUESS" (and please do take that definition literally) was based on what I saw occur, nothing more.  I wasn't theorising anything, as I don't have the appropriate level of technical expertise to present a theory.  Perhaps Teracopy (the application) doesn't actually report correct transfer speeds?  I don't know. 

 

I fully admit I am no storage/networking guru, but seriously mate, you have some sort of god complex.  How about engaging in a civil debate instead of being a complete jerk?  We could have discussed this like adults and worked out what exactly was going on in order for me to see the screenshotted transfer speeds, yet instead you choose to speak to me like one would a dog.  (Actually I wouldn't even speak to my dog like that.)  I have seen you lord it over many people on this forum in many different posts and it's pretty bloody ordinary to say the least.

 

It's a real shame as I truly do like learning - but not from someone who hoards his knowledge and only drip feeds them bits and pieces to make himself feel superior.  A sad case indeed.  Feel free to ignore my post btw, I will be doing you the same courtesy for any further posts from you.

Start your own post for your education. This thread was about using 4 Port 1Gb/s or 10 Gb/s NIC. I feel the OP deserves a real answer, not guesses.

OK, since you want to load 10TB, disable parity and you can get better performance. You are still looking at several months.

 

Several months? Really?

 

The OP's estimate was months.

Doing this on a gigabit connection would take months, so I was looking for a faster solution.

So, even with parity disabled, yes the estimate is probably still valid. My experience with Crashplan does not allow me to change that estimate.

 

On the other hand, "out of the box", if the source(s) can write to the reiserFS directly (SATA), perhaps in parallel, the time required will be reduced.

 

Writing to a parity-protected drive, I see transfer speeds of 20+MB/s.  This is, pretty well, a sustainable rate.

 

At 20MB in one second, 10TB is going to take 140 hours - that is less than a week ... hardly several months .... or is my maths wrong?  However, you have the advantage over me since I have never used Crashplan - perhaps that is extremely inefficient, and will only run at less than 5MB/s?

 

Admittedly, that will nowhere near saturate a Gbit link but, for practical purposes in my pattern of usage, my 500GB cache drive effectively gives me a transfer rate which approaches the capacity of a Gbit connection, sustainable for the duration of any writes which I perform.

 

And you are still ignoring reads!  The original poster did not stipulate whether he required to improve write or read performance.  I still believe that with carefully chosen hardware, unRAID could saturate a Gbit link on reads which, for most applications, is the more important scenario.  It was you who muddied the waters by reference to write performance only.

And you are still ignoring reads!  The original poster did not stipulate whether he required to improve write or read performance.  I still believe that with carefully chosen hardware, unRAID could saturate a Gbit link on reads which, for most applications, is the more important scenario.  It was you who muddied the waters by reference to write performance only.

 

Sorry, the other half of the discussion is in the Crashplan thread where arcane has been working on getting things to work;

If all you want to do is dump data from a machine on your lan onto unraid then yes, crashplan isn't the easiest way to go about it.

 

Crashplan does, however, offer many more features and a much more complex 'infrastructure' for backups if you want it to use it.

 

 

Blu-ray is encoded at 25-35 megabits/sec, streams are 4-8 megabits/sec. Yeah, those are bits, not byte. You planning on 100 streams?

 

Let me know when you are short on read performance...

Blu-ray is encoded at 25-35 megabits/sec, streams are 4-8 megabits/sec. Yeah, those are bits, not byte. You planning on 100 streams?

 

I don't recall anyone stipulating that unRAID could only be used to store video streams.

 

unRAID holds my photographic library.  Fetching a set of high resolution raw images could certainly benefit from a higher bandwidth link

unRAID holds my photographic library.  Fetching a set of high resolution raw images could certainly benefit from a higher bandwidth link

 

You should run out and buy several 10Gbe PCIe cards

http://www.newegg.com/Product/Product.aspx?Item=N82E16833106044

and switch.

http://www.provantage.com/supermicro-sse-x24s~7SUP929Y.htm

 

Don't forget the cables, this place will let you specify length required.;

http://www.cablesondemand.com/category/SFP%2B+CBL/URvars/Catalog/Library/InfoManage/SFP+_CABLES_%28DIRECT_ATTACH%29.htm

 

That should get you very impressive speed for fetching high resolution raw images. You might consider interleaving the files on different drives to increase the speed.

 

40Gbe is available if you need a more impressive number.

Ethernet channel bonding is supported in unRAID v5.  It's not yet in the GUI.

unRAID holds my photographic library.  Fetching a set of high resolution raw images could certainly benefit from a higher bandwidth link

 

You should run out and buy several 10Gbe PCIe cards .....

 

You really are a prize prat, with a childish mind, aren't you?

 

I shall be ignoring all your posts from now on.  If you ever post anything of value, which seems unlikely on current performance, I will, unfortunately, miss it.

unRAID holds my photographic library.  Fetching a set of high resolution raw images could certainly benefit from a higher bandwidth link

 

You should run out and buy several 10Gbe PCIe cards .....

 

You really are a prize prat, with a childish mind, aren't you?

 

I shall be ignoring all your posts from now on.  If you ever post anything of value, which seems unlikely on current performance, I will, unfortunately, miss it.

 

Your words, "certainly benefit from a higher bandwidth link", so higher bandwidth link you got, complete with parts list. I disagree with your benefit calculation, but you got what you asked for.