This is InsaneGeek's Typepad Profile.
Join Typepad and start following InsaneGeek's activity
Join Now!
Already a member? Sign In
Recent Activity
Well think of it this way, in a hypervisor virtualized environment a guest is simply a process running. An os isn't really able to manage what happens within a process only what resources a process can be given. So in a virtual environment the most granular performance management you can get is for the entire VM. True within a VM you can then use traditional methods to manage scheduling, etc inside it. The core reason is this: today one cannot split a process and run it on different physical servers, a process runs on one physical server at a time (however in app space you could use some MPI within a VM to access addtional resources on another box). The only way for a hypervisor to control the underlying processes would a paravirtualized configuration with parts of the hypervisor running within a VM which everybody is pretty much running away from these days. There is no reason that you can't virtualize a system and put a print, mail & db within the same guest and run it fine. You can use the same workload tools within a VM as you have in the past; but there is a good chance that if you have a cluster of hypervizors you can't move the workload to other servers around as well. For good or bad, virtualization in general is about trying to push the max limits and reduce the idle time of assets. If you want your VM cluster to be able to move workloads around to maximize capcity, more smaller VM's allow that to be done (at the expense of more VM's to management, so pick your poison). Think of it in the context of storage: in my view it's easier to move a 100TB of storage from mirrored to raid5 when it's chopped up into 10x 1TB luns rather than a single 10x TB lun. If I have to move 10TB at one time I have to have a 10TB free to land it, whereas chunked into smaller pieces of 10x 1TB luns my free space requirement to move might be just 1TB (or less) as a lun is migrated off I can then use it's space for the next lun.
The place I work for does just this (content aggregation for education & research markets). We have been selling content to libraries for decades and now on the fringes are in a way competing with Google & Wikipedia. At this point we are still in a good position because as you mention the data that is publicly available for free off these sites is not always accurate. So we pay buy or pay royalties for correct and unique source data and charge a subscription for it. There is an interesting question as to: is there a similar tipping point with big iron unix vs Linux on it's "not as good but good enough", some people say yes it will get good enough for free public data that inaccuracies won't matter that much, others say no for reference data it is either 100% accurate all the time or not and the incentive for free data to be that accurate is cost prohibitive. We also are grappling with the concept of permanent archive data. I had pretty much the exact conversation a few months ago about maintaining data for hundreds of years, data that we might have received digitally how do you maintain a pristine copy of that data forever without it being horribly costly.
Toggle Commented Nov 2, 2010 on The Information Business at Storagebod's Blog
Late to the party but, I've got to say; "Dear god please *no*" on that suggestion. I have to already wade through vendors constantly putting up their own carefully manufactured benchmarks that have no basis in real life; to have to deal with them intentionally using benchmarks to show their competition in bad light... probably will break my will to want to deal with any vendor at all. I already have to constantly slap vendors around for putting benchmarks in front of me that show iops only performing within cache, or a single app, shortstroked, or whatever goofball way they have done it. Even worse it gets in my executive's hands who have no idea about what happens under the cover of a storage array other than "ooohh rack with blinky lights" so I then have to spend hours trying to explain why the benchmark has no real meaning in any real life scenario that would apply to us... arggg!!! I can only imagine the horror that would descend upon me if I had to additionally deal with execs dragging whitepapers down to me as to why vendor sucks, because some other vendor found a specific case which never would be done in real life and only applies in an engineered benchmark world.
Toggle Commented Jun 10, 2010 on They know Jack.... at Storagebod's Blog
This is great news. After talking with people at EMCWorld, etc I basically got the statement of "no that's going to be vmax only and won't come back to the DMX". I thought the fast v2 for DMX was great news but... having jumped on the thin-provisioning bandwagon so the fast for DMX is *excellent* (as we have lots of terabytes of non-thin storage that was older or not thin appropriate we still will be able to use fast on the DMX) To me the software changes for the thin is the big great news. Being able to drain freespace in the pool is excellent, excellent news and was the one paintpoint I ran into last year in trying to move 60TB of one thin pool raid type to another without scoring new storage. You called it seeminly minor... seeminly is very apropo, it would have been a major, major time saver for me last year that directly affects the business as I had to basically get all the luns unbound but the last few tens of gig out of the to finally get the first datadev free. At the expense of sounding like a fanboy, I love the fact that EMC is continuing to give love to the DMX line. To me this is further proof that EMC has changed some of their old ways and is looking to keep me happy as a customer. The things that EMC has recently given for free, continuing to add/back port features to old products (especially when there is a new replacement out), really make me think the old *Evil* Machine Corporation of old is no longer. As a customer please, please continue doing this; these are good foils that I can use to show value to executives who constantly are wondering why storage costs more than the 1TB usb drive at Best Buy.
1 reply
That was me being a bit more cheeky as I've gotten the pitch from EMC in the past about deduplication, performance, host spots, etc bad. Shake the trees a bit and see if you had something technically unique to change your previous arguments against dedupe or was more "now it's ok". I still am not really comfortable with deduplication, love thin provisioning but dedupe on primary storage arrays I haven't come to management comfort yet. I'm less concerned about performance, but more on the unexpected expansion of data. Somebody decides to upgrade all the guest OS's from RHEL4 to RHEL5, or from Win2003 to Win2008, etc. Expecations of space have dramatically changed (albeit for a short time until the post dedupe occurs again). If someone signs up for the 90% space reclamation pitch and actually uses that space... they are going to have a bad day, maybe not today, tommorrow, but sometime. And as times are tight and management starts squeezing, the business guys are going to demand the storage guys use that space they see unused on some lean six-sigma report somewhere (why dedupe it if you aren't going to give it back out). Thin provisioning does have some of a similar issue of variable space consumption, but I'd say that storage admins can be infront of that much easier, if I see my backing storage is approaching percent free and get more backing storage for it. If some system admin in a different country decides to push out a bunch of upgrades at 3am my time... I could be having a bad morning. The system admin wouldn't come talk to storage guy on their upgrade as the space consumed is the same, and I (the storage guy) don't want to be involved everytime somebody decides to change a file. I don't have any real technical method to see this expansion of space coming. If/when primary inline dedupe comes, I think some of these issues might be reduced... the post processing is where these issues become magnified. (note I'm using dedupe in my backup environment and very happy with it, so I'm not against it...)
1 reply
I'm a customer (of both EMC & NetApp) not a vendor... and as I posted on Evan's blog. I want the marketing guarantees to go away you are not solving any of my problems, in fact your carnival barker "guarantees" get in my way of having an actual good conversation with my business units. Rather than having a conversation about features and implementation methods that apply to my business needs... it's about a guarantee. Having a guarantee is not the problem but having a *useless* guarantee is. Who doesn't do non-mirrored protection, who doesn't do thin provisioning, who isn't using SATA drives? These guarantees are not unique to your architecture so in my view it’s pretty much an empty guarantee and if a vendor is getting into my environment by selling some executive a useless empty guarantee we've started on the wrong foot from square one. It's basically like telling me I should buy your car because you have a sign on your building that every car is guaranteed to come have four wheels... you know that BMW dealer down the street he doesn't put a guarantee on his sign... think about that why don't you. You do want four wheels right? If you'd never seen a car before and have no understanding of things (as the business often is) you'd probably think that's a heck of a guarantee; but the guy who's seen a car before would be heading for the door post haste. I hope you guys selling those guarantees really think about this... I know you really, really want to sell me your goods and in the ends the gimmick that will probably a number of units based on it to the uninitiated, but are you really sure that's how you want to start that relationship with me in IT?
InsaneGeek is now following The Typepad Team
Feb 2, 2010