Here we go with a comment on a Mike Workman’s post ( Pillar‘s CEO), I posted it on his blog too but sometimes he is not very open to discussion 😉

Ok, your company is suffering from this recession a lot, and you’re likely to have too much spare time to write FUD instead of going to customers and try to sell your arrays.

If Compellent has the worst product, why are they growing so much, while you aren’t? (I can assure you that Commpellent hasn’t a very good marketing dept.) Anyway, let me answer to your FUD:

About software RAID

I’ve been listening to this dispute, software vs hardware raid, since years. It is a very useless dispute!

Why companies like NetApp and Compellent grow so steadily if “software RAID” is really so bad? Because customers aren’t looking for hardware, they just look for ways to cut storage costs and to add more virtualization capabilities!

Compellent chose, years ago, to write their own software, call it firmware, microcode or whatever you like, on standard x86 servers, instead of harnessing proprietary (limited) hardware. And now big vendors like IBM, HP and EM are following the very same way Compellent lead! HP will deliver next generation EVA on standard Proliant x86 servers and HP is quite a bigger company than you!

There are big advantages in using standard hardware instead of developing a proprietary one: you have a small company and if you spend too much in hardware development, you’ll end up with no resource for software features. Moreover, if you fail in hardware design (i.e.: a buggy chip), it will be very hard to recover! How much does a dedicaded ASIC development cost, in comparison to the acquisition of a powerful standard intel Xeon CPU? How can you set up and mantain such a complex R&D department? The market says: NO WAY!

This is why Compellent delivers tens of features like data progression and you can’t. Want some more? Compellent has Remote replication (sync, async, and snapshots delivery), when will you be able to do the same?

As regards “software RAID” performance, I can show you many customers REAL workload reports (one more well implemented software from Compellent 😉 ) pointing out how good are the REAL performances of Compellent’s controllers and architecture as a whole!

Compellent can write RAID 10 LUNs on first tier disks (SSDs of FCs) and eventually migrate blocks of less used data to other tiers (SATA) and RAID Levels. This feature Compellent calls Data Progression is a clear boundary line among “traditional AKA old” storage products vs new ones. As you can see, EMC is striving hard to show they too put this kind of functionalities on their arrays (you likely heard about FAST). Compellent does not use this feature to cope with any lack of performances, but to supply customers with a great advantage. As soon as you are able to sport similar functionalities on your arrays we’ll may speak about performance advantages this feature brings to customers.

About Cache

Compellent has 512 MB of write cache and 3 GB of read cache. What’s the problem?

Cache is useful for sequential writes only, when you do fully random database workload, the storage cache isn’t so important. BTW, Compellent’s cache works quite differently than yours, it supports dynamic blocks size! (from 2KB up to 256 KB). This features leverages the capability Compellent has to understand the size of the block you are writing, in order to allocate just the space needed, i.e.: when you write a 4K block of cache on a Pillar array, the system allocates 64K! (60K of wasted space), with its 8GB cache and 4K DB blocks written, you will run out of cache after 128K writes, right the same number of writes you can do with Compellent’s 512MB cache and dynamic block size.

Moreover, it seems quite usual that cache is used to store other structures than actual data blocks (i.e.: bitmaps for snapshot). Compellent does not, what about Pillar?

Mike, you’d better behave yourself or Santa won’t pay you a visit if you continue to lie.


Merry Xmas Mike!