Comments on: Introduction to iSCSI

By: piavka

piavka — Wed, 30 Nov -0001 00:00:00 +0000

I can\’t believe that you mixed the iscsi initiator and target roles – this is shame to have such a dummy and miss informative article on linux-mag.

By: jsoda

jsoda — Wed, 30 Nov -0001 00:00:00 +0000

seriously? an article on iscsi that mixes up target and initiator? wow.

By: laytonjb

laytonjb — Wed, 30 Nov -0001 00:00:00 +0000

Mea maxima culpa. I always get the initiator and target backwards. I have fixed the text and will fix Figure 3 in a few hours.

I have also changed the wording in a few places where it wasn\'t as accurate as it should have been.

Jeff

By: viralnexxus

viralnexxus — Wed, 30 Nov -0001 00:00:00 +0000

I think Jeff did a great job on introducing me to iSCSI and SANS. Mistakes will always be made, but bashing him sarcastically instead of correcting him politely shows a serious lack of class. Just my opinion.

-Victor

By: normalex

normalex — Wed, 30 Nov -0001 00:00:00 +0000

100% agree with viralnexxus.

By: dotcomstu

dotcomstu — Wed, 30 Nov -0001 00:00:00 +0000

What happens if the SAN dies?

This setup would benefit from having a DRBD network mirror to remove the single point of failure introduced by having all servers depending on the one SAN machine: http://www.drbd.org/

By: chammitt

chammitt — Wed, 30 Nov -0001 00:00:00 +0000

also, you would typically want to have a separate Ethernet Network (for SAN) from your traditional Ethernet Network (for other IP traffic) for performance reasons…not just one Ethernet Network. While possible to vlan it out…if you have decent any amount of workload, your performance will suffer. But then again, if you are planning to use 2TB+ SATA drives…this might not be an issue for you. I\’d stick to the 300-450GB SAS drives personally. spindle > capacity less you are doing a huge sequential workload….which most people do not!

By: zodilib

zodilib — Wed, 30 Nov -0001 00:00:00 +0000

I liked the article. However would love to know more about the initiator and adaptor story. Meanwhile please let me know if the network is geographically scattered can we still have a FC NAS setup ? My question must be dump,, forgive me I am just starting :)

By: laytonjb

laytonjb — Wed, 30 Nov -0001 00:00:00 +0000

@dotcomstu,

If the SAN dies, then you lose storage. Using DRDB is an option, but it\'s basically having a second network. So you could have a second SAN and achieve similar results (price and performance are TBD). But it is an option and an interesting one.

Jeff

By: laytonjb

laytonjb — Wed, 30 Nov -0001 00:00:00 +0000

@chammitt,

A very valid point about having 2 ethernet networks either physical or VLAN. I think there are two camps arguing this point. One camp wants a single big pipe for all traffic (call this the 10GigE camp) and the second camp wants a bit more redundancy and segregation of traffic. I don\'t think there is one right answer and depends upon the workload, the risk, and the desire (or need) for redundancy.

Personally, while the single (fat) line is very appealing, I would also like to have at least 1 GigE line to each server for other tasks (monitoring, imaging, etc.). As you point out, it helps reduce traffic on the iSCSI line. GigE isn\'t expensive so I don\'t see it being a problem overall. Plus, you could do multi-pathing across both lines recognizing the difference in performance.

Jeff

By: laytonjb

laytonjb — Wed, 30 Nov -0001 00:00:00 +0000

@zodilib,

The story behind initiator and target is that I got them backwards in the original post of the article. I think I fixed everything.

To be honest, I don\'t know enough about FC to know if you can run it over long distances. I\'m assuming there are companies that have WAN FC products (there are companies that do this for InfiniBand so I\'m assuming FC is the same).

Jeff

By: piavka

piavka — Wed, 30 Nov -0001 00:00:00 +0000

@viralnexxus,

IMHO mixing such fundamental things as iscsi initiator and target (not just in accidentalt in one place but overall in all article including the diagram) in 99% of cases indicates that the author is familiar with the subject only theoretically. I think that anyone who have actually made the simplest iscsi setup even just once, would not mix such things - but there can be exceptions of course.

My point is that I expect the author of the article not to be a novice in a subject he/she is writing about and at least have some real practical experience with the subject, also considering that it\'s posted not on authors personal blog but in a well known magazine where people reading it have certain expectations from the articles. And I don\'t mind if the article is introduction, intermediate or expert level - as long as it is fundamentally correct, while minor mistakes I\'m fine with of course.

I\'m sorry for such a response ,I should have been much more polite,
but honestly I never remember myself of ever writing such a bashing comment before.

By: bobgjunga

bobgjunga — Wed, 30 Nov -0001 00:00:00 +0000

What about the difference in latency between FC and GigE? 10 GigE? I think that bandwidth determines how many severs you can have on a SAN without performance degradation but latency limits the maximum single server performance.

So what is the performance comparison of having a MySQL data folder on..

a local 15k rpm SAS dirve

a FC SAN drive

a 15k rpm SAS dirve over a 1 GigE iSCSI SAN

--BobG