I believe there's a misunderstanding of A-SIS technology and how it is implemented based be discussions around performance impacts and concerns. A-SIS was designed to NOT be a real time de-dupe specifically to not have negative performance impacts. Instead A-SIS is a scheduled process which runs in a batch - style mode, allowing customers to schedule windows of operation around the demand cycles of their data sets.
WE should really move forward here. May I suggest that if you are considering deploying VMware on NetApp and leveraging A-SIS, you should ask to speak with reference accounts who use A-SIS in their production environments. I'd suggest that you'll be quite surprised. From there, I know you'll validate the technology i your lab, and eventually roll it into production.
For NetApp A-SIS is not new technology, rather its another use of NetApp pointer based file system WAFL. For 15 years NetApp has been able to manipulate data in manners not available to other vendors. From zero performance impact Snapshots, to SnapMirror, SnapVault, FlexVols, FlexClone, etc to the latest functionality A-SIS. If you like the previous technologies, then you'll like A-SIS.
Vaughn
Milazzo Giacomo wrote:
Hi All
sorry but I disagree with the cited "minimal system impact" that Vaughn cites talking about A-SIS. When we think to dedup processes in a Nearstore VTL appliance I can also accept the high overhead on CPU that this causes because in a "backup" environment we can also accept that this process slows the general I/O values but, as long NetApp offer the choice in the FAS appliances and their NAS usage (http, ftp, nfs, cifs) the dedup processes causes a tremendous impact on performances! In my experience we've tested and sold both the worst Quantum (formerly ADIC) dedup appliances, NetApp with A-SIS and Falconstor but the leader remains, for the moment of course, Datadomain, both in terms of long term time space saving and performances. Dadadomain borns for "backup" (VTL a/o disk staging) and a dedup appliance, in my opinion, must do only this: to store backup data and save space.
Regards,
*Da:* owner-toasters@mathworks.com per conto di Vaughn Stewart *Inviato:* mar 30/10/2007 19.26 *A:* Fote_Philip@emc.com; jack1729@gmail.com; ggwalker@mindspring.com; jeremy.page@gilbarco.com; toasters@mathworks.com *Oggetto:* Re: A-SIS questions
Phil,
The market offers many data deduplication technologies, most of which run at the server level, may or may not dedupe the production data set, and typically incur a performance penalty when enabled.
A-SIS runs on the array, deduplicates the prodcution data set, and incurs minimal system overhead. in addition, de-duplicated data sets can be replciated in their deduplicated state providing storage savings on the prodcution and remote data sets.
As I know you are aware that with any technology one should should follow best practices to ensure optimal results. I was about to ask you about your business challenges, but I noticed that as your are with EMC so I'd suspect that your business challenges aren't focused on reducing storage costs.
Maybe we could find a customer who would like to compare the benefits of A-SIS vs Avamar in a head to head evaluation where we could share the results on toasters! Would you be interested in working together to pull this off?
Vaughn Stewart
On 10/30/07, Fote_Philip@emc.com Fote_Philip@emc.com wrote:
Just a general question on ASIS. How does it impact system performance? Is it enough so you need to plan around it?
-----Original Message----- From: owner-toasters@mathworks.com [mailto:owner-toasters@mathworks.com] On Behalf Of Jack Lyons Sent: Monday, October 29, 2007 8:02 PM To: Glenn Walker Cc: M. Vaughn Stewart; Page, Jeremy; toasters@mathworks.com Subject: Re: A-SIS questions
I agree that Vmotion and HA / DRS can handle most hardware issues, but there is an advantage to using MSCS. An application restart can happen in seconds or minutes. I can also force applications over to the other node to apply OS patches. We have several Virtual-Virtual clusters for test environment and several physical-virtual clusters for production.
Jack
Glenn Walker wrote:
I'm still trying to figure out why someone would use MSCS on ESX:
MSCS is primarily used for HARDWARE-LEVEL fault tolerance (it won't
help
if the application crashes, so much as if the hardware dies).
VMotion can be used in place of MSCS for hardware-level fault
tolerance,
IIRC, which negates the need for MSCS within a VM.
-----Original Message----- From: owner-toasters@mathworks.com
[mailto:owner-toasters@mathworks.com]
On Behalf Of M. Vaughn Stewart Sent: Saturday, October 27, 2007 5:08 PM To: Jack Lyons Cc: Page, Jeremy; toasters@mathworks.com Subject: Re: A-SIS questions
FlexClone will clone a datastore, for VM level cloning granularity you
gonna have to wait I think something is just around the corner.
As for MSCS you need RDMs as VMDKs are not supported with MSCS
RDMs can be either FCP or iSCSI VMDKs can be on NFS or VMFS (which is over FCP or iSCSI)
We are in the process of testing ESX on NFS. We have 2 out of 60 on NFS now. I just started thinking about using flexclones, is any one using flex clones with NFS for creating clones of VM's.
Also, from what I have read, to use MSCS I need to use an RDM from iscsi or FC luns...correct?
Thanks Jack
M. Vaughn Stewart wrote:
Jeremy,
A-SIS has a slight write performance penalty (a few points) and typically also sees a slight read performance gain (of the same amount), so in short customers who are using it love it. Now you don't use it everywhere, like on DB files.
As for VMware on NFS, enjoy your solution it rocks, don't take my word for it. Google VMware on NFS and sort by date, you'll have allot of reading to do. Quotes from executives at EMC and VMware just support what you are deploying. Make sure to see NetApp's TR3428 for deployment details.
See ya Vaughn
I have read some of the info on how A-SIS works but I have a couple
of questions:
- Assuming an average compression rate (say /home with random
docs
in it) how much of a performance hit does it impose 2) Does it change how much data is sent over the wire when you use SnapMirror/SnapVault? I.e. if I get 30% compression on a 1 gig snapshot do I send 1 gig or 700 mb to update my target?
Oh, and for the folks who helped earlier, we are moving our Oracle and ESX systems to NetApp, the pSeries will be fibre connected, at least at first but the ESX stuff is all going on NFS.
~Jeremy *__* This message (including any attachments) contains confidential and/or proprietary information intended only for the addressee. Any unauthorized disclosure, copying, distribution or reliance on the contents of this information is strictly prohibited and may constitute a violation of law. If you are not the intended recipient, please notify the sender immediately by responding to this e-mail, and delete the message from your system. If you have any questions about this e-mail please notify the sender immediately.
-- Vaughn Stewart