More on Linux and iSCSI [info, not flame :)]
So I thought I would inject some info into the discussion. :)
Oliver Xymoron (and others?) mentioned that one could do iSCSI in userspace. Well, Intel has code at
http://sourceforge.net/projects/intel-iscsi
Just looking at the todo and glancing at the code shows that it is far from optimal, but at the same time it is open source and a working starting point for anyone interested in optimizing it.
Jeff
Quoted 5 lines · Jeff Garzik
So I thought I would inject some info into the discussion. :)
Oliver Xymoron (and others?) mentioned that one could do iSCSI in userspace. Well, Intel has code at
http://sourceforge.net/projects/intel-iscsi
The included userspace server is largely proof-of-concept code and could do with a fair amount of rounding out. Things this could use to make it interesting:
- authentication
- run-time configuration
- ability to serve from files and block devices (MD, LVM, crypto-loop)
- ability to serve from /dev/sgX interfaces with native SCSI (tapes, CDRW..)
Don't know what the state of interop with other initiators is though.
I'll also point out that for many Linux<->Linux purposes, nbd is a workable substitute.
Signature · 1 line
"Love the dolphins," she advised him. "Write by W.A.S.T.E.."
Quoted 16 lines · Oliver Xymoron
Quoted 5 lines · Jeff Garzik
So I thought I would inject some info into the discussion. :)
Oliver Xymoron (and others?) mentioned that one could do iSCSI in userspace. Well, Intel has code at
http://sourceforge.net/projects/intel-iscsiThe included userspace server is largely proof-of-concept code and could do with a fair amount of rounding out. Things this could use to make it interesting:
- authentication
- run-time configuration
- ability to serve from files and block devices (MD, LVM, crypto-loop)
- ability to serve from/dev/sgXinterfaces with native SCSI (tapes, CDRW..)
Quoted 4 lines
Don't know what the state of interop with other initiators is though.
I'll also point out that for many Linux<->Linux purposes, nbd is a workable substitute.
You are so correct in the Linux<->Linux model, but remember the other OS has the dominate market space. I have proof of interoperability. Better yet I have performance proof without the ability to control the benchmark environment. IIRC that reference is set at version 6.
The market space has version 7,8,9,11,12,13,16,18,19 working group varitions. So please anyone who wants to compete, I welcome the challenge. Do not forget the hardware, OS license investments.
Oh and Intel's Pro1000 T Storage Adaptor is a joy to work with, and you all can try to figure it out on your own. Oh and be sure the take notice the list of TOE's that do it sideways. All OS' mixing between ALL OS' is the challenge.
Do not forget the other dozen or so RFC docs to support the entire picture.
Cheers,
Andre Hedrick, CTO & Founder
iSCSI Software Solutions Provider
http://www.PyXTechnologies.com/
PS Thanks for the Advertising Minute!
Oliver et al.
http://downloadfinder.intel.com/scripts-df/filter_results.asp?strOSs=19%2C24%2C39&strTypes=DRV%2CFRM%2CUTL&ProductID=844&OSFullName=&submit=Go%21
http://downloadfinder.intel.com/scripts-df/proc/T8Clearance.asp?url=/4461/eng/Zama2_1.0.8_Linux_42715.tgz&agr=N
This self extracting file contains the firmware and software for the upgrading to 0.8 iSCSI specification.
I own this product, and have to install RH 7.1 w/ 2.4.2 kernels to use and test with it.
How about equal time for all?
Now to be fair, I am not looking to have Intel take any heat. I have already gotten abused over my issue, and it roles off.
Regards,
Andre Hedrick, CTO & Founder
iSCSI Software Solutions Provider
http://www.PyXTechnologies.com/
PS Thanks for the Advertising Minute!
PSS You are welcome Intel for the Advertising Minute!
Quoted 47 lines · Andre Hedrick
Quoted 16 lines · Oliver Xymoron
Quoted 5 lines · Jeff Garzik
So I thought I would inject some info into the discussion. :)
Oliver Xymoron (and others?) mentioned that one could do iSCSI in userspace. Well, Intel has code at
http://sourceforge.net/projects/intel-iscsiThe included userspace server is largely proof-of-concept code and could do with a fair amount of rounding out. Things this could use to make it interesting:
- authentication
- run-time configuration
- ability to serve from files and block devices (MD, LVM, crypto-loop)
- ability to serve from/dev/sgXinterfaces with native SCSI (tapes, CDRW..)Quoted 4 lines
Don't know what the state of interop with other initiators is though.
I'll also point out that for many Linux<->Linux purposes, nbd is a workable substitute.
You are so correct in the Linux<->Linux model, but remember the other OS has the dominate market space. I have proof of interoperability. Better yet I have performance proof without the ability to control the benchmark environment. IIRC that reference is set at version 6.
The market space has version 7,8,9,11,12,13,16,18,19 working group varitions. So please anyone who wants to compete, I welcome the challenge. Do not forget the hardware, OS license investments.
Oh and Intel's Pro1000 T Storage Adaptor is a joy to work with, and you all can try to figure it out on your own. Oh and be sure the take notice the list of TOE's that do it sideways. All OS' mixing between ALL OS' is the challenge.
Do not forget the other dozen or so RFC docs to support the entire picture.
Cheers,
Andre Hedrick, CTO & Founder
iSCSI Software Solutions Provider
http://www.PyXTechnologies.com/PS Thanks for the Advertising Minute!
Quoted 4 lines · Andre Hedrick
Quoted 2 lines · Oliver Xymoron
Quoted 1 line · Jeff Garzik
[..]
Quoted 7 lines
Quoted 4 lines
Don't know what the state of interop with other initiators is though.
I'll also point out that for many Linux<->Linux purposes, nbd is a workable substitute.
You are so correct in the Linux<->Linux model, but remember the other OS has the dominate market space.
Really? Must admit I haven't been paying attention to FreeBSD lately.. ;)
Quoted 2 lines
I have proof of interoperability. Better yet I have performance proof without the ability to control the benchmark environment.
I'm sure that's great, but until it's also DFSG compliant, it's effectively vaporware in my book.
Quoted 1 line
IIRC that reference is set at version 6.
They're actually up to 18.
Signature · 1 line
"Love the dolphins," she advised him. "Write by W.A.S.T.E.."
Quoted 11 lines · Andre Hedrick
Oliver et al.
http://downloadfinder.intel.com/scripts-df/filter_results.asp?strOSs=19%2C24%2C39&strTypes=DRV%2CFRM%2CUTL&ProductID=844&OSFullName=&submit=Go%21
http://downloadfinder.intel.com/scripts-df/proc/T8Clearance.asp?url=/4461/eng/Zama2_1.0.8_Linux_42715.tgz&agr=NThis self extracting file contains the firmware and software for the upgrading to 0.8 iSCSI specification.
I own this product, and have to install RH 7.1 w/ 2.4.2 kernels to use and test with it.
Which means, like every other binary module, it's pretty much worthless. Though I'm guessing that's not the point you're trying to make.
(For the record, there's not much value in iSCSI NICs, or TCP offload in general at the moment, except to avoid potential deadlock issues with trying to do network buffer allocation down in the block layer)
Signature · 1 line
"Love the dolphins," she advised him. "Write by W.A.S.T.E.."
Quoted 8 lines · Oliver Xymoron
Quoted 4 lines · Andre Hedrick
Quoted 2 lines · Oliver Xymoron
Quoted 1 line · Jeff Garzik
[..]
Quoted 2 lines
Quoted 2 lines
Don't know what the state of interop with other initiators is though.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
I'll also point out that for many Linux<->Linux purposes, nbd is a workable substitute.
You are so correct in the Linux<->Linux model, but remember the other OS has the dominate market space.
Really? Must admit I haven't been paying attention to FreeBSD lately.. ;)
Nah, it is Bill's and Scott's world.
See Enterprise storage is the name of the game. Ripping the heart out of the Fibre Channel Market is the goal. The Democratization of Storage is my goal.
Recall the humor of Penguin World Domination ? Well I am just paving the road for a place to Store it all !
Quoted 5 lines
Quoted 2 lines
I have proof of interoperability. Better yet I have performance proof without the ability to control the benchmark environment.
I'm sure that's great, but until it's also DFSG compliant, it's effectively vaporware in my book.
Oh that is as good as the uptime envy!
Thanks Google for the Cache and easy search!
Debian Social Contract
... lists) for non-free software packages. The Debian Free Software Guidelines (DFSG). Free Redistribution. The license of a Debian component ...
Description: License for Debian GNU/Linux.
Category: Computers > Open Source > Licenses www.debian.org/social_contract.html - 16k - Jan. 9, 2003 - Cached - Similar pages
Quoted 3 lines
Quoted 1 line
IIRC that reference is set at version 6.
They're actually up to 18.
That is nice to hear.
This project is Intel Lab's software reference implementation of iSCSI. *SEE README FOR COMPATIBLE VERSIONING IN EACH RELEASE*
* Development Status: 3 - Alpha
* Intended Audience: Developers
* License: BSD License
* Natural Language: English
* Operating System: Linux
* Programming Language: C
* Topic: System
That BSD only nicely keeps it out of Linux!
Regards,
Andre Hedrick, CTO & Founder
iSCSI Software Solutions Provider
http://www.PyXTechnologies.com/
Quoted 13 lines · Oliver Xymoron
Quoted 11 lines · Andre Hedrick
Oliver et al.
http://downloadfinder.intel.com/scripts-df/filter_results.asp?strOSs=19%2C24%2C39&strTypes=DRV%2CFRM%2CUTL&ProductID=844&OSFullName=&submit=Go%21
http://downloadfinder.intel.com/scripts-df/proc/T8Clearance.asp?url=/4461/eng/Zama2_1.0.8_Linux_42715.tgz&agr=NThis self extracting file contains the firmware and software for the upgrading to 0.8 iSCSI specification.
I own this product, and have to install RH 7.1 w/ 2.4.2 kernels to use and test with it.
Quoted 7 lines
Which means, like every other binary module, it's pretty much worthless. Though I'm guessing that's not the point you're trying to make.
(For the record, there's not much value in iSCSI NICs, or TCP offload in general at the moment, except to avoid potential deadlock issues with trying to do network buffer allocation down in the block layer)
Well if you restrict the transport of the protocol to on Network Fabric, you have a small point. However moving off the Network stack to MPI transports is real power of the protocol. Classic example is the IBM iSCSI Shark running Linux, and where are the their drivers? Recall they are the "Peace, Love, Linux" company. However, transports like Myrinet, SCI, Quads, etc ... are the place to be for local SAN banks.
When NIC's or TOEs support native ipsec, have a CRC32C offload core, and a few other issues them they will have value.
All the ones I have tested or seen, offload the network stack, do not support ipsec or handle the CRC32C offload.
Only the MPI fabric cards provide the tools desired.
Cheers,
Andre Hedrick, CTO & Founder
iSCSI Software Solutions Provider
http://www.PyXTechnologies.com/
Quoted 8 lines · Andre Hedrick
Oliver et al.
http://downloadfinder.intel.com/scripts-df/filter_results.asp?strOSs=19%2C24%2C39&strTypes=DRV%2CFRM%2CUTL&ProductID=844&OSFullName=&submit=Go%21
http://downloadfinder.intel.com/scripts-df/proc/T8Clearance.asp?url=/4461/eng/Zama2_1.0.8_Linux_42715.tgz&agr=NThis self extracting file contains the firmware and software for the upgrading to 0.8 iSCSI specification.
This has software updated to revision 18:
http://sourceforge.net/projects/intel-iscsi