Questions about write-back cache
#1
Posted 06 November 2012 - 01:05 PM
what my question is, is that some other diskless methods employs a write-back cache
1) i just want to understand what does a write-back cache do? i know it is used for iscsi diskless booting...
2) if i know the answer to 1) would it be possible to employ this into aoe?
3) do you know any other software like this to implement delayed caching to aoe? i'm thinking of deploying a single target in aoe to multiple workstations but it is said it can't be done due to the implementation of aoe...is this true?
Thanks for the enlightenment...
#2
Posted 06 November 2012 - 04:40 PM
Imagine that you have a hard disk drive that is somehow connected to multiple computers. Imagine that on that hard disk drive is a filesystem. Now imagine that the filesystem is not intended for simultaneous use by multiple systems. That is very similar to what you are talking about. Never use a single-computer filesystem with multiple computers at the same time if any of those computers are using the filesystem in a writable mode, or you will not enjoy the resulting corruption....i'm thinking of deploying a single target in aoe to multiple workstations but it is said it can't be done due to the implementation of aoe...is this true?
What you are probably more likely to enjoy is Copy-on-Write. Please see this article.
#3
Posted 06 November 2012 - 05:57 PM
although i'm new to grasp the concept of what you have given...
there are other iscsi diskless system that have this cache software on both clients and servers...i do think the general purpose of that is that write operations on the single target is synchronized, so that it does not trashes the filesystem on the example you provided...i know in fact that aoe alone doesn't have this because you send direct hdd parameters over the packet which vblade directly talks to the target...
if i would have to imagine what i like to do is that
aoe client driver -> vblade cache (if there's such a thing) - vblade
so that whenever multiple devices access a target, all write's are temporarily cached so that when it is finally full, the cache finally dumps to vblade, which i hope won't cause corruption...
okay, if that's not possible, how is that other diskless systems (that employ iscsi), be able to use a single target booting multiple clients?
btw, sha0 can you please be so kind to point me to a tutorial on how to use your winvblock...thanks...also nice work on that...
#4
Posted 06 November 2012 - 06:30 PM
You're welcome to it.thank you for replying sha0
I'm afraid that you are mistaken. iSCSI deals with disks, not filesystems....
there are other iscsi diskless system that have this cache software on both clients and servers...i do think the general purpose of that is that write operations on the single target is synchronized, so that it does not trashes the filesystem on the example you provided...
For example, you can have any disk filesystem you want to have, on a disk. You can even invent your own. Do you believe that some iSCSI product will know how to make sure your invented filesystem works nicely with multiple computers? Here is the problem: Suppose computer A reads the filesystem and says "the length of somefile.txt is 300 bytes." Then suppose computer B truncates somefile.txt so that it's only 42 bytes. Then suppose computer A tries to read the 300 bytes. This doesn't work. Delaying writes doesn't work either, since that results in a series of writes from multiple computers which do not make sense together.
iSCSI is similar. iSCSI uses IP. AoE does not.i know in fact that aoe alone doesn't have this because you send direct hdd parameters over the packet which vblade directly talks to the target...
I already mentioned the answer at the bottom of my first response. If you were thinking of a particular product, I suppose you'd mention it... So you must be asking about the general strategy, instead.if i would have to imagine what i like to do is that
aoe client driver -> vblade cache (if there's such a thing) - vblade
so that whenever multiple devices access a target, all write's are temporarily cached so that when it is finally full, the cache finally dumps to vblade, which i hope won't cause corruption...
okay, if that's not possible, how is that other diskless systems (that employ iscsi), be able to use a single target booting multiple clients?
I don't know of any tutorial for WinVBlock. Maybe I've forgotten.btw, sha0 can you please be so kind to point me to a tutorial on how to use your winvblock...thanks...also nice work on that...
#5
Posted 06 November 2012 - 06:48 PM
I already mentioned the answer at the bottom of my first response. If you were thinking of a particular product, I suppose you'd mention it... So you must be asking about the general strategy, instead.
right you are...okay, i'm not gonna mention any product...
okay let me cite another example...the use of gamedisk in diskless systems, the same gamedisk used by multiple workstations...
i mean if iscsi is similar to aoe in what you say (block device in windows), how come the concept of gamedisk that multiple workstation uses it (iscsi)? without any data corruption?
#6
Posted 06 November 2012 - 10:24 PM
okay let me cite another example...the use of gamedisk in diskless systems, the same gamedisk used by multiple workstations...
i mean if iscsi is similar to aoe in what you say (block device in windows), how come the concept of gamedisk that multiple workstation uses it (iscsi)? without any data corruption?
I already mentioned the answer at the bottom of my first response.
1 user(s) are reading this topic
0 members, 1 guests, 0 anonymous users