Sorry about the car accident DN. Here is Steve D., talking up iSCSI......
iSCSI and trends in the convergence of IP networking
When: Monday, October 23rd at 1:00 p.m. EDT (17:00 GMT).
Speaker: Steve Duplessie
Moderator: Welcome to our Real Time Vendor Connection with with Enterprise Storage Group analyst Steve Duplessie. Please ask your questions now.
sjacobs289938: What is iSCSI?
Steve_Duplessie: iSCSI is a proposed standard to do block data calls over IP based networks, specifically, Ethernet. It was started by Cisco and IBM and has received the most support initially. In my opinion, it will be the adopted standard, with some modifications and additions, but there is still lot's of work to do.
hmccarthy857229: Who are some of the players in the early market?
Steve_Duplessie: Nishan, NuSpeed (Cisco now), SAN Valley, CNT, Pirus, 3Ware, and SANcastle.
msavage531394: Isn't IP too slow for storage?
Steve_Duplessie: IP isn't the problem. TCP is the problem. It kills your host and slows down everything. People are trying to fix this by hardware assisted TCP NIC cards (Alacritech) and VI based NICs (Giganet, Troika). I've seen some (note: lab situation!) awsome performance out of a gig E networks running block and file data -i.e. like 99% of wire speed at sub 7% CPU utilization. If you could run block data at least as fast as on a separate fibre network over your existing infrastructure, why go with the separate net?
jobrien197702: What is SoIP?
Steve_Duplessie: SoIP is Nishan Systems "storage over IP" protocol. They are trying to get their protocol attached to iSCSI, believing that they have a more complete story to solving the problems associated with things like ordered messaging and dropped packets - the things TCP deals with.
rma859868: Why use IP vs. Fibre Channel for storage?
Steve_Duplessie: IP has lots of standards, Fibre doesn't. IP is necessary as a primary network, fibre isn't. If we only used file servers, we all could plop any box we wanted on the wire and have it speak standards (NFS, CIFS, TCP/IP, etc.). If it using the standard network had no performance ramifications, why wouldn't you?
epelczar394651: What is the point of storage over IP?
Steve_Duplessie: In my view, the reason we have all this noise about storage over ip, is basically due to the fact that in theory, if we could do "block" data calls over our existing network, we wouldn't need to create complex secondary storage networks, like a fibre channel SAN. Note the "theory" word here.
jtomasello66470: Would you deem lucent's optistar/edgeswitch to hold a technilogical advantage in the fibre/ip realm?
Steve_Duplessie: I like what Lucent has done with Vixel to create an IP/Fibre extension story. Sure beats the Cisco/Brocade proposed stuff that won't even be real for another 9 months or so. This is a good alternative to tie geographically disparte SANs together over an existing IP infrastructure. Advantage technically? Not sure, the proof will be in the pudding - who's selling what?
mthompson539734: In broad terms can/will iSCSI replace Fibre Channel or is this complimentary technology
Steve_Duplessie: One can make a legit case that ASSUMING (note the tone here people) that all the technical challenges are worked out with iSCSI, why would you want a separate network (fibre channel)? Don't get me wrong - users HAVE to use fibre today to solve real world problems, but if all the hype turns to reality, we could very well come to a point whereby fibre becomes irrelevent. years away, however, from even being possible.
Moderator: We are still taking questions. Please submit any questions in the narrow, horizontal rectangle at the bottom of the chat screen. Then, press the Enter key to send it. Don't worry, if Steve doesn't get to your question we will add it to transcript that will be posted on the site within 48 hours after this event.
Lloyd.Dyrow43167: What will iSCSI offer over NAS protocols?
Steve_Duplessie: Not the same thing! iSCSI is for "block" data, NAS is "file" data. While both can and will exist over a common IP network, block vs. file will always exist. There will be other ways to do block data over IP a la pseudo drivers, and there will also be ways to do block within a file (at least it will look to the driver like block calls).Cool stuff.
kevin.parrish676275: Are there any Storate Area Network companies working with iSCSI, currently?
Steve_Duplessie: Not sure what you mean, but 3Ware has an iSCSI array already (as far as they can, iSCSI isn't done yet). Adaptec and others are shipping IP/Storage solutions. most of the large array vendors are on board (because why wouldn't they?). The only ones dead against it at this point are the fibre purists.
jobrien115602: What are the advantages of using IP with SANs?
Steve_Duplessie: (Theory) one unified network. I have to have IP, I don't "have" to have fibre channel, or any other secondary storage network. One thing is easier to deal with than two. IP is everywhere, compatibility isn't an issue, standards abound, and life's little problems have all been worked out. An independent storage network requires new and different stuff, management, etc. and all that costs money.
bruce.fitch157293: How will the proposed storage over IP protocol maintain data packet sequence order?
Steve_Duplessie: In a nutshell, that's the problem. It still hasn't been determined, but I think they will continue to use TCP to provide sequence ordering.
michael329906: Please explain iSCSI technology, and how it differs from implementations such as NFS and CIFS like what Network Appliance is using?
Steve_Duplessie: iSCSI is for "block" data - i.e the raw mechanism for scsi drivers accessing a disk. NetApp and others in the NAS world don't deal with blocks (on the front end), they deal with FILES. NFS,CIFS, FTP, etc. are all standards based on file data. Files are easier to manage than blocks from a human perspective, but many applications like to talk to blocks vs. files. The lines will get fuzzier as companies like NetApp provide file advantages to block applications. Thats a good question, must be a vendor.
charlie.cassidy311874: Could you compare iSCSI with Infiniband and Fiber Channel?
Steve_Duplessie: Fibre channel and Infiniband are examples of a transport - i.e. the wire type, so to speak. iSCSI is a protocol, that can run over those wires, or Ethernet wires. They don't really compare. SCSI is both a transport and a protocol, which most people screw up. With fibre channel, we really speak SCSI (protocol) over Fibre Channel (transport). Now I bet your confused.
thoyt898203: Is iSCSI a feasible solution currently?
Steve_Duplessie: No. It will be, but it's way to early in the game to be used in any sort of mission critical space. Expect cicso/nuspeed to offer first gen. products in Q1, but that will be for extended distance usage, not to replace data center SANs.
weintraub_jean980730: Do you think that iSCSI will replace fibre channel implementations? If so, what timeframe do you expect?
Steve_Duplessie: I think ETHERNET based storage solutions, of which iSCSI will be a part, (VI, DAFS, etc. all will play a role) will be prime time capable within 24-36 months, if all goes well. I don't think anyone will "replace" fibre channel that is working and is implemented, but moving forward in 3-5 years, they may not continue with fibre channel. This shouldn't seem so revolutionary, even DEC moved from CI to DSSI to SCSI sooner or later (lot of good that did them, however).
sjacobs289938: What did NuSpeed have that Cisco wanted?
Steve_Duplessie: NuSPeed proved that a disk drive could be viewed, and used, by a server as a local SCSI device, even though it was really connected via Ethernet. The ramifications are obvious - more stuff over IP is good for Cisco, so storage represents a huge potential upside for them.
charlie.cassidy311874: If TCP is the porblem, wouldn't a solution like SoIP or Adaptec's storage transport be a better solution?
Steve_Duplessie: Maybe. CNT has had a working solution longer than anyone. The problem is, TCP is religion. I don't think anyone will ultimately win that war unless Cisco says so. Better doesn't necessariliy mean thats what wins (for example, Hitachi has had the best disk array on the planet for years, and I think they sold 6 of them). iSCSI is backed by the big dogs, and those are the folks that will decide ultimately.
henrik.kraft792884: Why are we seing storage over IP emerging; isn't Fibre Channel good enough?
Steve_Duplessie: Good enough? what is that? what is ever good enough???? Fibre is hard. SAN is brutal. Stuff doesn't work together, it's expensive, and we sold the user community a bill of goods. The poor user has enough trouble keeping a float, and now we tell him he needs to have an entirely separate network? That ain't good enough.
henrik.kraft792884: What did Cisco acquire in NuSpeed?
Steve_Duplessie: 62 hockey players and -73% winters, along with the ability to push 10X the traffic over IP.
henrik.kraft792884: What are the technologies emerging connecting SANs over WANs / MANs?
Steve_Duplessie: The first real benefit to IP/Storage is tying together disparate fibre SANs via IP (ethernet). That's what San Valley, NuSpeed (kind of), Nishan, Lucent/Vixel, Brocade/Cisco, etc. are doing. Let's you at least view/use geographically (MAN) dispersed SAN "islands" as a universal SAN. cool stuff.
charlie.cassidy311874: Will most people put storage on a separate backend LAN, or will they put it on their regular IP network?
Steve_Duplessie: I'd separate it, but with switching technology as it is, it may become a moot point.
ssvitenk971772: So if you do not use TCP, and with the products you have seen, speed of IP is close to that of fiber, or other transmission mediums?
Steve_Duplessie: I've seen TCP (hardware assist) numbers on NetApps DAFS and Giganet NICs using VI that run at 99% of wire speed with sub 7ms CPU overhead. To me, thats the end game.
Lloyd.Dyrow43167: Will iSCSI be secure?
Steve_Duplessie: That's certainly one of the issues involved. security is still going to be based at the host, so that won't change. If someone can hack your system today and get to block data, the same will be true in iSCSI.
stpierre_edgar592673: Are there any optimizations or changes required to TCP in order for iSCSI to be feasible? And if so, what are they and how far out are they?
Steve_Duplessie: Not at face value. but, TCP is the pig that kills systems, so hardware assisted TCP is mandatory in my view. This isn't new, Alacritech is doing it, and everyone on the planet has a project going on in the space. It will become the norm, iSCSI or not.
jnoh878245: Steve, can you please comment on the impact of FCIA's recent 10gig announcement?
Steve_Duplessie: My answer is "whoopdi do". Bout time. soon the fibre community will be able to agree on lunch. The point is there is no point. This is a press release, not much more. I hope they can make the stuff work, cause users who are either in now, or will buy in over the next few years, will need to get to 10G just like the Ethernet guys will. The fibre guys have been talking 10G for years.
Moderator: We are still taking questions. Please submit any questions in the narrow, horizontal rectangle at the bottom of the chat screen. Then, press the Enter key to send it. Don't worry, if Steve doesn't get to your question we will add it to transcript that will be posted on the site within 48 hours after this event.
mcbosetti531490: Would you explain what differentiates iSCSI from SCSI?
Steve_Duplessie: SCSI uses a standard SCSI driver in the OS to communicate directly to a disk (tape) reading and writing "blocks" of data to/from the disk. Ethernet based storage traditionally is file based, hence the term file server. iSCSI is the combination - use the scsi driver and make it look like the disk is attached directly to the host server, but in reality it's connected somewhere on the iP network.
kevin.parrish676275: Is iSCSI a concept, or a working prototype? Are there any RFCs? What is the current status or state of iSCSI?
Steve_Duplessie: It's a working group. there are working prototypes. it will be another year before it's done, me thinks.
khaller915360: Isn't it true that storage has always had a separate network because I/O was too slow over a network? Therefore, it isn't just block data holding back IP networks but that storage I/O typically needs a dedicated channel - like Fibre Channel, ESCON, etc.
Steve_Duplessie: Yes. storage causes TCP to kill IP networks, traditionally, therefore we've always needed a separate bus/network. that's what everyone is trying to fix.
wade_hoffman469926: Of the companies working on various IP products, are any of them dealing with the 'glass-house' data traffic issues such as redundant paths, delivery guarantees, etc.?
Steve_Duplessie: CNT has it now, but there own implementation of TCP. Pirus will have a real high end offering within 6 months. You could argue that Lucent also has carrier class gear today.
Moderator: We are still taking questions for the next five minutes or so. Please submit any questions in the narrow, horizontal rectangle at the bottom of the chat screen. Then, press the Enter key to send it. Don't worry, if Steve doesn't get to your question we will add it to transcript that will be posted on the site within 48 hours after this event.
ross_paul450498: In an earlier answer, you mentioned technical challenges with iSCSI. Could you elaborate on what are the top challenges that need to be worked out?
Steve_Duplessie: Ordered messaging and TCP are the biggest challenges. i.e. IP can deal with dropped packets or wrong sequences, storage can't.
michael329906: You mentioned "file" vs. "block" with iSCSI and NAS. Is the technology mature enough to contrast performance between the two technologies? Should block theoretically be faster?
Steve_Duplessie: It always was faster, but the lines are getting closer and closer. With VI, file can be faster than block.
wade_hoffman469926: Of the various IP storage networking proposals like iSCSI, etc., what, if any are the advantages, disadvantages.
Steve_Duplessie: Doesn't matter (though a good question). as soon a Cisco bought NuSpeed, iSCSI won.
jerry.zeisler22630: How will the storage controllers deal with the latencies? How well storage performance across a long distance IP network?
Steve_Duplessie: WAN latencies will be a problem whether it's iSCSI or fibre channel - it wont matter. If the application cant deal with WAN latency, it will crap out regardless of protocol.
Steve_Aucoin550105: What technology break-through is needed to make this really take off, and how long do you think it will take for this to become reality?
Steve_Duplessie: I don't think it really is going to be a tech. breakthrough - I think it's going to be Cisco kicking butt and making $$$ happen. The technical always has a way of working itself out once enough dollars are at stake.
s_desai787370: What are some of the early applications that will use iscsi
Steve_Duplessie: All of them. that's the beauty, apps won't even know (unless their is vast latency). To an app, it's talking to a SCSI disk as always.
nik.simpson978269: Do you foresee a time when a NAS device is just another client connected to a SAN rather than having its own dedicated storage.
Steve_Duplessie: Beautiful. Arguably, that's what we do most of the time. NetApp is the NAS king but uses fibre channel on the back end and will have a switched SAN sooner or later. EMC plugs Celerra into it's Sym in a SAN. So yeah, the question is will the SAN always be fibre or will it also be Ethernet? ATL just introduced an Ethernet tape library, which is totally cool, whereby users connect via standard ethernet and each DLT has an NDMP server on it.
mattaliano_paul980129: Is this similar to EMC's SRDF over IP?
Steve_Duplessie: SRDF is an application that runs over IP, or Escon.
mark23642: Who is the leader in driving the iSCSI standard?
Steve_Duplessie: 1. Cisco. 2. See one.
cardillo_john29335: How will iSCSI fit into backup solutions, tape or disk?
Steve_Duplessie: Not sure it will. ATL already has a net connected library, where each DLT has an NDMP server on it, so users can back up direct from machine to tape over ethernet. may make sense, but may not be necessary.
mparenti593520: Which SCSI over IP protocol do you think will survive (SoIP, ISCSI, SEP, ...)?
Steve_Duplessie: iSCSI.
steve.ainsworth823992: Can the current IP (ipv4 based) infrastructure/network support the throughput demands of storage traffic or will re-engineering be required?
Steve_Duplessie: I'm not smart enough to answer that.
joe_strong35455: A lot of the arguement for using IP networks as a transport for data access that I hear is that the customers already have an infrastructure built around IP, and they wouldn't have to build a seperate SAN infrastructure. I think this ignores the fact that a lot of IP networks are overloaded right now. How many customers really have a dedicated Gigabit or other private network that would be suitable for this type of traffic?
Steve_Duplessie: True, but the point is users will still HAVE to get an updated IP network, but do they HAVE to get another fibre network?
Moderator: This Real Time Vendor Connection with Steve Duplessie will end in 5 minutes. Unfortunately we have a backlog of questions, so we have to stop accepting them. Steve will answer all the questions we have accepted so far and the answers can be found in the transcript that will be available within 48 hours.
dana_richard820797: Hadn't CNT already proved this point?
Steve_Duplessie: I'm sure they have proved many points. Which one are you referring too? The point of no return? Point blank? Pointelism? You get my point.
markb85627: How does the Cisco/nuspeed solution address performance / response times over WAN?
Steve_Duplessie: It doesn't. it was a connection only. they are working with others to mitigate performance issues associated with IP/WAN.
kenn70173: With disks being seen as an IP device, are there going to be any distance limitations or latency issues?
Steve_Duplessie: Yes. applications that require certain response times will not be able to use remote iSCSI devices (at least as they are now) as primary storage. Localized, it will end up being the same.
rkugele912942: Is Storage Over IP a threat to SSPs using Fibre Channel?
Steve_Duplessie: No. SSPs, or anyone for that matter, need to take care of todays storage problems today - and that means fibre and it means SAN. As the technologies evolve, all enterprises will be afforded more choices, which is never a bad thing.
khaller915360: Are the blocks that you mention pieces of a database or blocks on a disk drive? If either of these, wouldn't other vendors like Oracle have to be involved to make this work?
Steve_Duplessie: Could be either. Oracle won't have to do anything for iSCSI specifically, because they, and all apps, see iSCSI as SCSI, so they still deal with raw blocks.
leonard_knapp38191: With 10Gig FC coming, will that make a difference in the FC/Ethernet debate, or will it be too expensive?
Steve_Duplessie: All I know is I can buy a 5 port 100Mb switch for about $39 bucks at CompUSA, and it's $3 grand a port for fibre. 10G aint going to make it cheaper.
bruce.fitch157293: Cisco is the network; how does IBM add value/benefit in this space?
Steve_Duplessie: IBM, as goofy as they can be, is a huge innovator and massive storage player.
steve_pheneger15530: What are the expected implications to storage outsourcing utilities and centralize data management with the convergence toward IP?
Steve_Duplessie: None. Storage is still a science, as well as an art, and talent is non-existent. IP won't fix that so I remain hugely bullish on the SSP sector.
ed.herlihy715689: Steve, You say that SAN is hard, and that isn't good enough . What about the storage virtualization products? don't they address the "stuff dosen't work together" issue?
Steve_Duplessie: They do, and I love them. I wish they existed 2 years ago and we wouldn't be in this mess. I think it's important for the whole fibre community to hop behind the storage virtualization folks, then everyone will A: sell more stuff and B: help the poor user. This space is brand spankin new though, so people don't even know about it. It wasn't around when folks were ripping their hair out last year.
tony.funari98282: Which companies in this space have the best chance of succeeding?
Steve_Duplessie: Cisco, because they are Cisco, but I like Pirus, San Valley, CNT actually makes money (so they may be doomed!), I love Nishan, but they sort of offended both Cisco and Brocade (why didn't they call EMC bozo's while they were at it?). 3Ware is first with a disk array based on iSCSI. Hell, there's lots of folks that sound good, but it's just too early to tell.
seshadri_raj473647: What kind of response time(MB/sec.) we are talking about with iSCSI?
Steve_Duplessie: I've seen it run at near wire speed, 1Gb/second. I've also seen it run like crap.
mkachmar25310: What happen to companies like Brocade / McData products if iSCSI become dominate ?? Why offer SANs when LAN technolgy is similar ??
Steve_Duplessie: I think brocade will hop into the iSCSI mix sooner or later. They have no real competition for the next few years, so don't sell the stock, but sooner or later they will have to either join the fray, or justify why not. Fibre purists are the ones who will suffer eventually.
jviviano172201: Regarding latency, will be going back to the days where 30ms response times were sufficient and cached arrays were not yet prevelant?
Steve_Duplessie: We're an immediate gratification society. but the bottom line is 90% of the applications we use we couldn't care less about 100ms response times. I run word and excel, and couldn't tell 100ms from 1ms. There are some apps in OLTP where it definetely does matter, but the lions share of real world apps don't have this concern any longer.
Moderator: Unfortunatley, we have to end our Real Time Vendor Connection with Steve Duplessie. Steve has given us some great insight on iSCSI and trends in the convergence of IP networking. Thanks very much for all of your great questions. We also want to thank Steve for his time. All questions that have not been answered will be in the transcript that will be posted within 48 hours. |
