Dell Buys IPO-Bound EqualLogic for $1.4 Billion
alphadogg writes "Dell is stretching further behind PCs and servers and boosting its storage business with a $1.4 billion buyout of EqualLogic, a storage company that filed to go public in August. CEO Michael Dell had hinted just last week that Dell could be on the prowl for some big game."
Translation: They're all overpriced and the market is a bit overheated.
I also wonder, who's using these storage companies? Is it for backups of corporate data centers?
I prefer Flambe as apposed flamebait.
Dell is just preparing to launch the biggest SAN-based Halo 3 server ever.
I wonder if this has any implications for Dell's partnership with EMC. Will Dell not be pushing EMC's low-end iSCSI storage now that they have their own? Or do the offerings from this new acquisition not compete at the same level as the EMC products?
Disclaimer: I work at EMC, but have no inside knowledge concerning Dell or this acquisition.
OpenFiler.
Fantastic piece of kit.
Website Hosting
EqualLogic was kind of a cool company that bundled value and decent software engineering into a good package, and had good support for stuff besides just Windows and Linux (VMware, NetWare, etc). Good service, etc. There are probably more than a few EqualLogic customers that are less than thrilled about this.
I've been talking to Equallogic guys at various VMWare events and I think they've got a great product. The biggest pain with implementing ESX has been the cost around shared SAN storage (yes you can use NAS now but come on.)
I love VMWare, but can't stand their parent corp (EMC) and can't wait for the Compaqification of the SAN market with the part of IBM played by EMC. Any company that forces their customers to buy $100 SATA drives for $900 deserves to die at the hands of commoditization.
Fsck the millennium, we want it now.
Millennium Crisis Line: 0890 900 2000 [calls cost 50p/min]
The whole EMC thing makes me wonder the reason for the purchase. That having been said, we've been using EqualLogic for 2 1/2 years, and at the price point, it's pretty much unbeatable. It's ridiculously easy to configure and grow. We've a small staff with a huge amount of storage (35 TB) due to the image intensive nature of our work (digital imaging). EqualLogic has been a life save for us from both time and performance standpoints. We could not have gotten purchase approval for anything faster.
EqualLogic seemed to have a relatively well integrated hardware and software solution as well as a good support system. I wonder how Dell taking the reigns will affect this as we were considering implementing a EqualLogic in our organization.
More info on Forbes.
EqualLogic is a data storage systems provider based in Nashua, N.H., with over 3,200 customers in 30 countries. Its virtualization products allow a single computer to function like multiple machines, so companies can spend less on hardware and energy costs in data centers.Eh? Aren't they talking about VMware here? EqualLogic sells storage solutions. If he means that several machines can access those storage systems simultaneously, he's out on a limb! I've never heard of such systems!
I spent last weekend running a number of performance tests (using IOMeter) against a couple of iSCSI targets on a oldish server and there's a performance issue that I just don't understand. The initiator is MS's software on W2K3 R2. I hooked up five 10k spindles and a dedicated RAID box (RAID0) to the server's U160 SCSI HBA. Running IOMeter locally on the server, performance was pretty good for the hardware, about what I'd expect -- with four threads & 4k chunks & 100% random & 50% R/W, an avg req time of 7 mS and a few hundred IOPS. Using the Starwind target with a small RAM disk to measure the raw end-to-end iSCSI performance the numbers were very good (avg req time of 1-4 mS), regardless of the network.
But When I "marry" the storage to the iSCSI target software, instead of getting req delays of A+B = 7+4 = 11 mS, I see delays of 22 mS. With the same target hardware running Openfiler the avg delay was 26 mS.
Very interestingly, the speed of the network had almost no impact on performance unless I was testing with large chunks and sequential reads. With random I/O, I got essentially the same performance using 802.11g wireless as I did using 1GBit hardwired! The RAID had 128MB of writeback cache and was using cached I/O. The drives were using command tag queuing.
Why oh why is there such an added "penalty" latency when combining the storage with the iSCSI target application? I was testing with consumer network gear at home, but I got about the same numbers using a dedicated HP commercial switch at work. Also roughly the same numbers when using an LSI RAID card in the server instead of the external RAID. I want to build a moderate performance iSCSI SAN for use with a couple of Exchange 2007 boxes, but I'm not going to bother if I have to take a huge penalty hit for using iSCSI instead of cobbling up some local storage.
You don't mention the switches you're using, but from the fine Equallogic training class I went to a couple of weeks back, a couple of the big things that can effect iSCSI performance are flow control on both the switch and client's ethernet interfaces, disabling unicast storm control on the switch and turning on jumbo frames. The general drift was that iSCSI can pound the snot out of switches and cheap ones just don't work all that well. Stacked Cisco 3750s are apparently popular, and there were a couple comments that (ahem) Dell switches suck for iSCSI.
Also, what makes you so sure Dell is would stop selling fibre channel? Or what makes you think Dell would stop reselling EMC fibre channel devices?
iSCSI is so... wrong for a SAN. The logic is it _can_ be cheaper because you can reuse existing network equipment. *cough* And skip the iSCSI HBAs. *ahem* and single path. Yaaaayyy, now it's just block-level NAS.
My guess is Dell made this decision to compete with HP for price sensitive customers looking at server/storage bundles.
I work for a company that sells these equalogic units (and install/train) and they are quite impressive.
I wont bang on about them, but they have a number of plus's which make them great for smaller companies and they scale quite remarkably well. They're pretty simple to manage and so forth (but you can get at the guts of them, they run a bsd variant).
One of the things i do appreciate about them is its 1 cost and you get everything they have to offer (on the software side anyways) and you dont have to get any golden screwdriver/license upgrades.
Performance wise in the bang for buck arena they are very competative though.
The biggest problem with selling them? iSCSI. Alot of people just dont seem to get iSCSI at all and write it off as a "less secure and slow" fibre channel option. We sell fibre channel products as well, and while fc has its advantages, they really aren't in the area of speed or latency (often thats true even when your not using a iSCSI hba).
But, as i said, i work for a reseller who stocks these and am ultimately going to be written off as biased (and yes we sell 3par, hds, sun and so forth as well).
Thanks for the reply. My puzzlement is that when measuring the pure iSCSI performance (with a ramdisk on the target) performance was great, regardless of the network -- avg I/O delays of 1mS with one IOMeter worker thread and 4mS with four worker threads... and a few hundred IOPS. This is all random I/O and the total throughput in MB/s is fairly low and not important for my app.
When the iSCSI target was disconnected from the ramdisk and connected to the real drives, the avg I/O delays were considerably worse than just adding the pure iSCSI overhead (1-4 mS) and the local performance on the target machine (7 mS). Instead of about 8-11 mS I get an avg I/O delay of 22-26 mS.
This is a simple "SAN" with one initiator and one target, no MPIO, etc. For pure random I/O there was no performance difference between a dedicated HP ProCurve, consumer 100MBit, consumer 1GBit, and consumer 802.11 wireless! There was a major performance difference between the networks with sequential I/O and large blocks of course.
Please write these test and your methodologies up and posit it somewhere. You can even probably get some ad revenue by doing that. I'm sure you'll also get a lot os suggestions as to what the trouble might be. I know I'd like to see the full details of each test myself.