SCI 28Apr2015 Latest ESRP (Exchange 2013) performance results analysis

In PowerEdge R730xd, PowerVault MD1200, PowerVault MD1220 by AdministratorLeave a Comment

This Storage Intelligence (StorInt™) dispatch covers Microsoft Exchange 2010 (E2010) and Exchange 2013 (E2013) Solution Review Program (ESRP) results.[1] Since we last reviewed this over 5000-mailbox category nine months ago, there have been at least five new submissions, two from Dell (PowerEdge R730xd, at 8K & 12.5K mailboxes), one from Fujitsu (DX200 S3, at 20K mailboxes) and one from HP (3PAR StoreServ 7400c [converged], at 40K mailboxes).

Latest ESRP V3.0 & 4.0 (E2010&E2013) performance

In Figure 1 we show our top ten database backup MB/server chart.

SCIESRP150428-001Figure 1 SCI’s Top ten ESRP database backup MP per second per server

Both the new Dell R730xd server-storage systems show up well here (@ #1 & #4). These Dell solutions incorporate an Exchange server and storage together, using internal as well as external disk shelves for storage. Similar configurations were used for the Dell PowerEdge R720xd and Dell Powervault MD1200 & MD1220 submissions. All these used SAS connected internal and external disk storage with Dell servers running the Exchange benchmark.

We don’t show this chart very often, as it is a per server perspective and many of the over-5K mailbox ESRP submissions use multiple Exchange servers connected to one storage system. Although this is not as much of an issue when you have servers directly connected to the storage. Instead, we typically show a total database backup (across all servers) chart but none of the new submissions performed well enough to be included there.

Although, the new Dell systems also performed well on Exchange backup throughput per database, we like that chart even less as there’s always multiple databases that reside on a storage system. Rather than including that chart, in Figure 2 we turn to a scatter plot of DB transfers per second per disk spindle split up by RPM.

SCIESRP150428-002Figure 2 SCI’s Scatter plot ESRP database transfers per second against disk spindle count

We have shown this chart before but this incorporates the latest results. As one can see in Figure 2, 15Krpm disks continue to do best (@~47.2 database transfers/sec/spindle) on standard database transfer activities, with 10Krpm (@~41.8 database transfers/sec/spindle) and 7.2Krpm (@~31.7 database transfers/sec/spindle) disks coming in 2nd and last place respectively. The correlations (for the linear regressions) vary from ~0.50 to ~0.91, which indicate for 15Krpm and 7.2Krpm there was some variability in performance per spindle across all submissions.

Presumably, the heavy randomness and corresponding high seek activity of Exchange database transfer activity favors 15Krpm disks. This is probably due to its better rotational latency, great seek times and relatively small data transfer size. This chart and the two that follow exclude any results that use flash storage or flash cache.

In Figure 3 we see a similar plot but this time for total database backup in MB/second.

SCIESRP150428-003Figure 3 SCI’s Scatter plot ESRP total database backup MB/second against disk spindle count

In Figure 3, 10K rpm (@~17.5MB/sec/spindle) and 7.2K rpm (@~16.8MB/sec/spindle) disks do much better than the 15K rpm disks (@~8.6MB/sec/spindle) and correlations are worse (ranging from -0.003 to +0.220). Such low correlations indicate a relatively high amount of variance for MB/sec per disk spindle on this workload.

It’s hard to explain what we see here, other than perhaps the 7.2K and 10K rpm do better on sequential throughput where seek speed and rotational latency are not as important. Also, higher capacity per track may help here as well, because this would lead to less seeking. Nevertheless, we have always been under the impression that 15Krpm disks had better throughput than 10K or 7.2Krpm disks. Apparently, there’s more to Exchange database backup IO than throughput alone.

Finally, in Figure 4 we show another version, only this time we are plotting log throughput in MB/sec against spindle count, segmented by disk RPM.

SCIESRP150428-004Figure 4 SCI’s Scatter plot ESRP log throughput MB/second against disk spindle count

In Figure 4, we see an even more interesting view, with 10Krpm doing best (@~5.3 KB/sec of log throughput/spindle), 15Krpm disks coming in 2nd (@~3.1 KB/sec of log throughput/spindle) and 7.2Krpm disks last (@~1.1 KB/sec of log throughput per second). Correlations are much worse here, ranging from ~-2.04 to ~+0.15, indicating even more variability in log throughput per disk spindle.

Unclear just what this is trying to tell us other than log processing throughput is a combination of database transfer IO as well as backup-like, sequential throughput activity. There also may an impact from newer disk drives (10Krpm) have more recent and capable controllers which process diverse IO faster than older controllers. The advent of flash storage is driving 15Krpm disks out of the market and as such, they haven’t been updated as often and may be falling behind the technology curve.

Summary

We find the scatter plots of performance per drive illuminating, to say the least. Here we can clearly see that the various ESRP benchmark activity (database transaction, database backup and log processing) represent three distinct IO workloads. The fact that 15Krpm drives don’t perform best on all of them is a bit of a mystery. Mostly, we would attribute their mixed performance as indicating that 15Krpm drives do poorly in long sequential transfers and aren’t being updated as much anymore. The fact they did do very well in database transfers indicates there’s still a place for 15Krpm disk in randomized, short block, OLTP workloads.

Any constructive criticisms on how to improve our analyses for any of our performance reports are always welcome. Moreover, if you detect errors in this or any other of our performance reports please do let us know and we will correct it as soon as possible.

If you received this report from someone else, please consider signing up for your own free copy of SCI’s StorInt™ monthly newsletter using the QR code below right (click on subscribe to newsletter upper right). We provide commentary on recent major storage system announcements as well as analyze current storage system performance results/benchmarks like this each month. We plan to analyze Microsoft ESRP results again in three months.

[Also we offer more block storage performance information plus our OLTP, Email and Throughput ChampionsCharts™ in our recently updated (November 2018) SAN Storage Buying Guide, or for more information on some select ESRP performance results please see our recently updated (October 2018) SAN-NAS Storage Buying Guide, both of which are available for purchase on our website.]

[This performance dispatch was originally sent out to our newsletter subscribers in April of 2014.  If you would like to receive this information via email please consider signing up for our free monthly newsletter (see subscription request, above right) and we will send our current issue along with download instructions for this and other reports. Dispatches are posted to our website at least a quarter or more after they are sent to our subscribers, so if you are interested in current results please consider signing up for our newsletter.]  

Silverton Consulting, Inc., is a U.S.-based Storage, Strategy & Systems consulting firm offering products and services to the data storage community

[1] ESRP results from http://technet.microsoft.com/en-us/exchange/ff182054.aspx, as of 27Apr2015

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.