State of Cloud Storage Providers Industry …...á Peer1 Hosting (powered by EMC ATMOS) á Rackspace Cloud It is worth noting that EMC Atmos Online originally passed performance, stability
Post on 22-Jul-2020
3 Views
Preview:
Transcript
State of Cloud Storage Providers Industry Benchmark Report: A Comparison of Performance, Stability and Scalability
Executive SummaryCloud storage is a young industry, but one that clearly has great promise for growth. Large organizations such as
1HWøL[��$PD]RQ��*RRJOH��1%&�DQG�=\QJD�DOUHDG\�GHSHQG�RQ�FORXG�VWRUDJH�WR�PDQDJH�DQG�UXQ�WKHLU�EXVLQHVVHV��
and cloud storage providers have only just begun to serve smaller enterprises. The unique architecture of the
cloud not only offers unlimited storage capacity, but also lays the groundwork for eliminating the daily grind
of data backup thanks to the cloud’s constant replication of data. As a result, this new market is populated by
many competing cloud storage providers (CSPs) offering varying degrees of performance and reliability. Since
no two CSPs are created equal, potential customers need to identify those that can deliver the appropriate
SULFH�SHUIRUPDQFH�OHYHOV�WR�PHHW�WKHLU�VSHFL÷F�EXVLQHVV�QHHGV��EXW�IHZ�SRVVHVV�WKH�WLPH��UHVRXUFHV�RU�LQ�KRXVH�
H[SHUWLVH�WR�SHUIRUP�D�WKRURXJK�DQG�SURDFWLYH�HYDOXDWLRQ���
In April 2009, Nasuni began an ongoing and unprecedented evaluation
process of CSPs based on a wide array of factors. Nasuni offers enterprise
FXVWRPHUV�DFFHVV�WR�LWV�6WRUDJH�6HUYLFHV��ZKLFK�FRPELQH�RQ�SUHPLVH�
security and snapshot technologies with the vast raw storage resources
RI�VRPH�RI�WKH�OHDGLQJ�&63V��%\�EHLQJ�&63�DJQRVWLF��1DVXQL�FDQ�VHOHFW�
IURP�DPRQJ�WKH�EHVW�SURYLGHUV�IRU�LWV�FXVWRPHUV��%XW�WR�GHWHUPLQH�ZKLFK�
SURYLGHUV�ZHUH�WKH�EHVW�÷W�WR�PHHW�VSHFL÷F�QHHGV�1DVXQL�FRPPLWWHG�WR�
completing rigorous CSP testing – a process that took more than two
years and is still ongoing. It provides an unprecedented view of the real
performance of the cloud. Indeed, these tests enabled Nasuni to
correctly predict that two CSPs would likely soon leave the market due
to poor performance.
7KLV�WHFKQLFDO�WHVW�ORRNHG�DW�KRZ�&63V�SHUIRUP�LQ�D�VSHFL÷F�XVH�FDVH��
QDPHO\��KRZ�WKH\�SHUIRUP�IRU�PLG�VL]HG�RUJDQL]DWLRQV�WKDW�ZDQW�WR�WDNH�
advantage of the cloud for primary storage, data protection and disaster
UHFRYHU\��6RPH�SURYLGHUV��IRU�H[DPSOH��KDYH�EXLOW�WKHLU�FORXGV�WR�RSWLPL]H�
WKHLU�DELOLW\�WR�GHOLYHU�YHU\�ODUJH�PHGLD�÷OHV�DFURVV�WKH�:HE��1DVXQL�GLG�
not test for every possible use case.
$V�WKH�÷UVW�RI�LWV�NLQG��WKLV�UHSRUW�UHYHDOV�WKH�÷QGLQJV�RI�WKLV�RQJRLQJ�WHVWLQJ�DQG�HYDOXDWLRQ�SURFHVV��IRFXVLQJ�RQ�
three key areas of importance for organizations that want to take advantage of cloud storage:
Page 1 / www.nasuni.com / info@nasuni.com
2011 © Nasuni Corporation, All Rights Reserved
+1.508.433.6200 / +1.800.208.3418 International US and Canada only
$V�WKH�÷UVW�RI�LWV�NLQG��
this report reveals
WKH�÷QGLQJV�RI�WKLV�
ongoing testing and
HYDOXDWLRQ�SURFHVV��
focusing on three key
areas of importance for
organizations that want
to take advantage of
cloud storage.
Performance
The cloud needs to respond
quickly to queries and not slow
when stressed. If performance is
poor, organizations will spend too
much time waiting on the cloud
and productivity will suffer.
Stability/Availability
If organizations are to trust
critical data to the cloud, it must
be available at all times.
Scalability
One of the primary advantages
of cloud storage is unlimited
FDSDFLW\��:LWKRXW�WKLV�SURSHUW\��
the cloud is much less valuable
to organizations.
8OWLPDWHO\��RQO\�VL[�RI����SURYLGHUV�SDVVHG�1DVXQLÚV�WHVWLQJ�
á� Amazon S3
á� AT&T Synaptic Storage as a Service (powered by EMC ATMOS)
á� Microsoft Azure
á� Nirvanix
á� Peer1 Hosting (powered by EMC ATMOS)
á� Rackspace Cloud
It is worth noting that EMC Atmos Online originally passed performance, stability and scalability testing.
However, EMC discontinued this service in 2010 and now sells EMC Atmos as a platform solution that enables
RWKHU�RUJDQL]DWLRQV�WR�KRVW�WKHLU�RZQ�FORXGV��%DVHG�RQ�DGGLWLRQDO�WHVWLQJ��WKH�TXDOLW\�RI�VHUYLFH�IURP�WKRVH�ZKR�
use ATMOS to provide CSP services varies a great deal from provider to provider.
Two CSPs emerged as top performers in the Nasuni study: Amazon S3 and Microsoft Azure, with Amazon S3
being the standout across all evaluation areas.
MethodologySince April 2009, Nasuni has conducted ongoing tests of the 16 largest CSPs. These tests are tiered so CSPs must
SDVV�WKH�÷UVW�WHVW�EHIRUH�SURFHHGLQJ�WR�WKH�QH[W�WHVW��7KH�÷YH�WHVWLQJ�VWDJHV�DUH�DV�IROORZV�
á� API Integration, to ensure that it is possible to test the service at all. Nasuni wrote a custom connector in
Python for each provider according to the documentation of the CSP’s API and available best practices. The
test itself was run through Python’s own unit testing framework to provide pass/fail results for each of the
basic APIs.
á� Unit Testing, in which larger software components are broken down into their building blocks (units) and
then tested for inputs, outputs and error cases. This enables Nasuni to determine whether each CSP can
VXFFHVVIXOO\�KDQGOH�EDVLF�IXQFWLRQV�VXFK�DV�ZULWLQJ�DQG�UHDGLQJ�GLIIHUHQW�÷OH�VL]HV��SURYLGLQJ�DFFHVV�IURP�
multiple locations simultaneously, and writing to the cloud with one connection and then reading the same
÷OH�EDFN�IURP�D�GLIIHUHQW�FRQQHFWLRQ�
Page 2 / www.nasuni.com / info@nasuni.com
2011 © Nasuni Corporation, All Rights Reserved
+1.508.433.6200 / +1.800.208.3418 International US and Canada only
Two CSPs emerged as top
SHUIRUPHUV�LQ�WKH�1DVXQL�
study: Amazon S3 and
Microsoft Azure��ZLWK�
Amazon S3 being the standout
across all evaluation areas.
á� Performance Testing, to measure response time (how quickly one can interact with the cloud), throughput
(how fast data can move back and forth to the cloud), and the impact of a higher level of stress. To do this,
Nasuni imposed a number of variables:
á� Concurrency (multiple simultaneous threads),
á� Object size (ranging from very small to very large), and
á� Workload type��UHDG�YV��ZULWH��DQG�PL[HV�
In performance testing the cloud, location can bias or affect the
provider being tested. Tests on Amazon S3 from Amazon EC2 yield
better results than Amazon S3 tests from Rackspace Cloud Servers,
IRU�H[DPSOH��7R�PLQLPL]H�ORFDWLRQ�ELDV��1DVXQL�UDQ�EHQFKPDUNV�IURP�
three different geographic areas and hosting companies using the
power of cloud computing. In addition, some cloud storage
SURYLGHUV�KDYH�RSWLPL]HG�IRU�ODUJH�÷OHV�DQG�VPDOOHU�REMHFW�FRXQWV��
ZKLOH�RWKHUV�KDYH�H[FOXVLYHO\�IRFXVHG�RQ�VPDOOHU�÷OHV��1DVXQLÚV�
REMHFW�VL]H�WHVWV�HQDEOHG�LW�WR�÷QG�WKH�UDUH�IHZ�WKDW�SHUIRUP�ZHOO�DW�
both ends of the spectrum.
á� Stability Testing,�WR�DVVHVV�WKH�ORQJ�WHUP�UHOLDELOLW\�RI�HDFK�&63��1DVXQL�SHUIRUPV�WKLV�WHVW�WKURXJK�WKH�
IROORZLQJ�SURFHVV��ZKLFK�LV�UHSHDWHG�HYHU\�÷YH�PLQXWHV�RQ�DQ�RQJRLQJ�EDVLV��
1. :ULWH�D�÷OH�WR�WKH�FORXG�
2. ,QIUHTXHQWO\�UHDG�EDFN�D�÷OH�IURP�WKH�FORXG�IURP�D�SUHYLRXV�ZULWH��QRW�WKH�PRVW�UHFHQW�RQH��DQG�
FRPSDUH�LW�DJDLQVW�H[SHFWHG�GDWD��DQG
3. 'HOHWH�DQ�ROG�÷OH�IURP�WKH�FORXG�
7KH�UHVXOWV�RI�WKLV�WHVW�UHYHDO�UHVSRQVH�WLPH�FRQVLVWHQF\�DQG�WKH�IUHTXHQF\�DQG�GXUDWLRQ�RI�&63�GRZQWLPH���
1DVXQL�PHDVXUHV�ORQJ�WHUP�UHOLDELOLW\�IURP�VHYHUDO�ORFDWLRQV�VR�WKDW�UHVXOWV�DUH�QRW�FRORUHG�E\�ORFDO�
connectivity issues.
7R�SDVV��WKH�SURYLGHU�KDG�WR�SHUIRUP�ZLWK�QR�GDWD�ORVV�DQG�KDYH�QR�VLJQL÷FDQW�XQSODQQHG�RXWDJHV��DOWKRXJK�
SODQQHG�RXWDJHV�E\�&63V�DUH�DQ�LQGXVWU\�FRQFHUQ��(QWHUSULVHV�VKRXOG�H[SHFW�&63V�WR�EH�DYDLODEOH������RI�
the time with no planned outages.
á� Scalability Testing, to understand how well each CSP handles high object counts. Different clouds have
GLIIHUHQW�GH÷QLWLRQV�RI�ÜKLJK�REMHFW�FRXQWV�Ý�,Q�DQ�LGHDO�ZRUOG��1DVXQL�ZRXOG�FRQGXFW�DQ�HQGOHVV�WHVW�WKDW�
ZULWHV�DV�PDQ\�÷OHV�RI�GLIIHUHQW�VL]HV�DV�LW�FRXOG�WR�WKH�FORXG�LQ�SHUSHWXLW\�DQG�ZRXOG�PHDVXUH�ZULWLQJ�
UHVSRQVH�WLPH�WR�GHWHUPLQH�ZKHWKHU�LW�GHJUDGHV�DV�WKH�V\VWHP�÷OOV�XS��7KH�LGHDO��RI�FRXUVH��LV�QRW�IHDVLEOH��
since most CSPs charge per operation, and testing costs would quickly become unsustainable.
7R�SDVV��WKH�VHUYLFH�FRXOG�
not break under testing.
$GGLWLRQDOO\��SURYLGHUV�
failed if they requested
WKDW�1DVXQL�FHDVH�WHVWLQJ�
because the test was
overstressing their cloud.
Page 3 / www.nasuni.com / info@nasuni.com
2011 © Nasuni Corporation, All Rights Reserved
+1.508.433.6200 / +1.800.208.3418 International US and Canada only
)RU�LWV�WHVWLQJ��1DVXQL�GH÷QHV�ÜVFDODELOLW\Ý�EDVHG�RQ�WKH�IROORZLQJ�
1. +RZ�PDQ\�REMHFWV�RQH�FDQ�SODFH�LQ�WKH�FORXG�
2. :KHWKHU�SHUIRUPDQFH�UHPDLQV�FRQVWDQW�DV�RQH�÷OOV�WKH�FORXG�RU�LI�LW�GHJUDGHV�RYHU�WLPH��and
3. The reliability of the cloud over a long period of sustained high operation rates.
)RU�WKLV�WHVW�1DVXQL�ZURWH�VPDOO�÷OHV���.%��ZLWK�PDQ\�WKUHDGV�RI�FRQFXUUHQF\�IRU�VHYHUDO�ZHHNV�RU�DW�OHDVW�
until it hit 100 million objects. As Nasuni did so, it measured and tracked error rates and performance.
To pass, the CSP had to cross the 100 million object threshold successfully without degradation.
Results
Basic Functionality: API and Unit
Nasuni’s testing began with 16 CSPs. The initial qualifying tests set out to determine:
1. Whether it is possible to build a connector for the service
(API testing); and
2. Whether the service can perform basic functions, such as
ZULWLQJ�GLIIHUHQW�÷OH�VL]HV��VLPXOWDQHRXV�DFFHVV��DQG�ZULWLQJ�WR�
WKH�FORXG�ZLWK�RQH�FRQQHFWLRQ�DQG�WKHQ�UHDGLQJ�WKH�VDPH�÷OH�
back from a different connection.
7KH�÷UVW�WZR�VWDJHV�RI�EDVLF�WHVWLQJ�×�$3,�DQG�8QLW�×�ZHHGHG�RXW�PRUH�
WKDQ�RQH�WKLUG�RI�SURYLGHUV��2QO\�RQH�IDLOHG�DW�WKH�$3,�VWDJH��WKRXJK�
VHYHUDO�RWKHUV�UHTXLUHG�PXOWLSOH�SDVVHV�DQG�H[WHQVLYH�GLVFXVVLRQV�RYHU�
several months with technical support to produce a working API.
$W�WKH�XQLW�OHYHO�WHVWLQJ�VWDJH��÷YH�RI�WKH����UHPDLQLQJ�&63V�IDLOHG��1DVXQL�HQFRXQWHUHG�PLVVLQJ��LQFRUUHFW�RU�
LQDSSURSULDWH�GRFXPHQWDWLRQ��DQG�RIWHQ�D�VLPSOH�ZULWH�UHDG�GHOHWH�RSHUDWLRQ�ZRXOG�IDLO��HVSHFLDOO\�LI�TXLFNO\�
H[HFXWHG��0RVW�IDLOXUHV�ZHUH�UHODWHG�WR�WKH�GLVWULEXWLRQ�DQG�DFFHVV�RI�FRQWHQW�ZLWKLQ�WKH�&63��,I�WKH�&63ÚV�
DUFKLWHFWXUH�ZDV�GHVLJQHG�IRU�VORZ�DFFHVV�ZLWK�OHVV�WUDQVDFWLRQDO�RSHUDWLRQV�VXFK�DV�DUFKLYLQJ�÷OHV�RU�VDYLQJ�
photos, Nasuni’s tests would severely stress the system, sometimes to the breaking point.
Performance
$W�WKH�SHUIRUPDQFH�OHYHO��1DVXQL�H[DPLQHG�KRZ�IDVW�&63V�FDQ�ZULWH�DQG�UHDG�÷OHV�LQ�D�YDULHW\�RI�VL]HV��2I�WKH�
���SURYLGHUV�WKDW�TXDOL÷HG�IRU�WKLV�VWDJH�RI�WHVWLQJ��WZR�IDLOHG�GXH�WR�SHUIRUPDQFH�OHYHOV�WKDW�ZHUH�WRR�ORZ�
WR�EH�DFFHSWDEOH�WR�WKH�YDVW�PDMRULW\�RI�HQG�XVHU�RUJDQL]DWLRQV��8QOLNH�WKH�$3,�DQG�XQLW�WHVWV��KRZHYHU��WKH�
SHUIRUPDQFH�WHVW�GRHV�QRW�JHQHUDWH�D�FOHDU�FXW�SDVV�IDLO�UHVXOW��'HSHQGLQJ�RQ�KRZ�D�FXVWRPHU�PLJKW�PDNH�XVH�RI�
the cloud, a lower performance level may be acceptable. Caching, for instance, can shield an end user from slower
SHUIRUPDQFH�WLPHV�ZKHQ�UHDGLQJ�÷OHV��
$W�WKLV�OHYHO��1DVXQL�was establishing which CSPs have the most basic capabilities required to provide services to customers.
Page 4 / www.nasuni.com / info@nasuni.com
2011 © Nasuni Corporation, All Rights Reserved
+1.508.433.6200 / +1.800.208.3418 International US and Canada only
7KH�UHVXOWV�UHSUHVHQWHG�LQ�WKH�IROORZLQJ�FKDUWV�VKRZ�WKH�SHUIRUPDQFH�RI�WKH�WRS�VL[�SURYLGHUV�ZLWK�ODUJH�÷OHV����0%���PHGLXP�VL]HG�÷OHV������.%��DQG�VPDOO�÷OHV����.%��
Charts 1a and 1b show results
from performance tests on
ODUJH�÷OHV��ZLWK�VSHHGV�VKRZQ�LQ�
megabytes per second. In terms
RI�ZULWLQJ�ODUJH�÷OHV�WR�WKH�FORXG��
Microsoft Azure had the highest
DYHUDJH�VSHHG�DW������0%�VHF��
ZLWK�1LUYDQL[�FORVH�EHKLQG�DW�
�����0%�VHF��7KH�UHVW�RI�WKH�WRS�
VL[�SURYLGHUV�SRVWHG�FRPSDUDEOH�
VSHHGV�ZLWK�WKH�H[FHSWLRQ�RI�
Peer1, whose average large write
VSHHG�ZDV������0%�VHF�
)RU�UHDGLQJ�ODUJH�÷OHV��1LUYDQL[�
ZDV�IDVWHVW�DW������0%�VHF��
followed by Microsoft Azure at
�����0%�VHF��$PD]RQÚV�DYHUDJH�
VSHHG�WRRN�WKLUG�ZLWK�������0%�
sec and Peer1 once again fell far
EHKLQG�WKH�UHVW�ZLWK�����0%�VHF��
less than a quarter the speed of
the top two.
Please note that on charts 1a and 1b,
larger bars represent faster speeds.
Page 5 / www.nasuni.com / info@nasuni.com
2011 © Nasuni Corporation, All Rights Reserved
+1.508.433.6200 / +1.800.208.3418 International US and Canada only
&KDUW��E��$YHUDJH�5HDG�6SHHG�IRU�/DUJH�)LOHV����0%�
&KDUW��D��$YHUDJH�:ULWH�6SHHG�IRU�/DUJH�)LOHV����0%�
&KDUW��D��$YHUDJH�:ULWH�6SHHG�IRU�0HGLXP�)LOHV������.%�
&KDUW��E��$YHUDJH�5HDG�6SHHG�IRU�0HGLXP�)LOHV������.%�
Charts 2a and 2b compare
KRZ�TXLFNO\�HDFK�RI�WKH�WRS�VL[�
performers could read and write
PHGLXP�VL]HG�÷OHV������.%���
)RU�ZULWLQJ�PHGLXP�VL]HG�÷OHV��
Microsoft Azure was fastest with
DQ�DYHUDJH�VSHHG�RI�����0%�VHF��
followed closely by Amazon S3 at
����0%�VHF��7KH�RWKHU�SURYLGHUV�
IHOO�VLJQL÷FDQWO\�EHKLQG�WKH�WRS�
two, with most performing 28 to
41 percent slower than Microsoft
Azure, and with Peer1 performing
70 percent slower than the leader.
Concerning read speeds for
PHGLXP�÷OHV��$PD]RQ�ZDV�
the clear leader, far ahead of
HYHU\RQH�HOVH�DW�����0%�VHF��,Q�
second place, Microsoft Azure
was 28 percent slower at 6.6
0%�VHF��$77��1LUYDQL[�DQG�
Rackspace posted comparable
VSHHGV�LQ�WKH�����WR�����0%�VHF�
range, with Peer1 again coming in
ODVW�DW�����0%�VHF��OHVV�WKDQ�RQH�
quarter the speed of Amazon.
Please note that on charts 2a and 2b,
larger bars represent faster speeds.
Page 6 / www.nasuni.com / info@nasuni.com
2011 © Nasuni Corporation, All Rights Reserved
+1.508.433.6200 / +1.800.208.3418 International US and Canada only
&KDUW��E��$YHUDJH�:ULWH�6SHHG�IRU�6PDOO�)LOHV����.%�
&KDUW��D��$YHUDJH�5HDG�6SHHG�IRU�6PDOO�)LOHV����.%�Charts 3a and 3b compare how quickly each of the top 3 performers could read and write VPDOO�÷OHV����.%��LQ�WHUPV�RI�WKH�number of objects per second. ,Q�UHDGLQJ�VPDOO�÷OHV��$PD]RQ�S3 far outpaced its rivals, with ����÷OHV�VHF�����SHUFHQW�IDVWHU�WKDQ�LWV�FORVHVW�ULYDO��$77��3HHU��was once again the laggard at 46 ÷OHV�VHF�����SHUFHQW�VORZHU�WKDQ�Amazon S3.
In terms of performance for writing objects, Microsoft Azure WRRN�WKH�WRS�VSRW�ZLWK�����÷OHV�sec, closely followed by Amazon 6��DW�����÷OHV�VHF��$77�FDPH�LQ�WKLUG�DW����÷OHV�VHF��ZKLOH�WKH�RWKHU�WKUHH�×�1LUYDQL[��5DFNVSDFH�and Peer1 – fell far behind with VSHHGV�RI����÷OHV�VHF�����÷OHV�VHF�DQG����÷OHV�VHF�UHVSHFWLYHO\�
Please note that on charts 3a and 3b,
larger bars represent faster speeds.
2YHUDOO��$PD]RQ�DQG�
0LFURVRIW�$]XUH�FRPH�
out as the strongest
providers when it comes
WR�SHUIRUPDQFH��ZLWK�IDVW�
read and write speeds
across the board. For
those end users who are
mostly concerned with
VWRULQJ�ODUJH�÷OHV��KRZHYHU��
1LUYDQL[ÚV�QHWZRUN�
SHUIRUPHG�H[WUHPHO\�ZHOO�
Page 7 / www.nasuni.com / info@nasuni.com
2011 © Nasuni Corporation, All Rights Reserved
+1.508.433.6200 / +1.800.208.3418 International US and Canada only
Stability7R�GHWHUPLQH�D�&63ÚV�VWDELOLW\��1DVXQL�FUHDWHG�MREV�WKDW�ZURWH�D�÷OH�WR�WKH�FORXG��LQIUHTXHQWO\�UHDG�WKH�÷OH�EDFN�WR�
FRPSDUH�LW�DJDLQVW�H[SHFWHG�GDWD��DQG�WKHQ�GHOHWHG�DQ�ROG�÷OH�IURP�WKH�FORXG��7KH�JRDO�ZDV�WR�GHWHUPLQH�
1. +RZ�RIWHQ�WKH�FORXG�JRHV�GRZQ�
2. +RZ�ORQJ�GRZQWLPH�ODVWV�
3. +RZ�ORQJ�LW�WDNHV�WR�H[HFXWH�WKLV�WHVW�
4. +RZ�WKH�WLPH�LW�WDNHV�WR�H[HFXWH�WKH�WHVW�FKDQJHV��DQG
5. :KHWKHU�GDWD�ZULWWHQ�FDQ�DOZD\V�EH�UHDG�EDFN�VXFFHVVIXOO\�
The purpose of this test is to determine whether CSPs can deliver the
FDOLEHU�RI�UHOLDELOLW\�UHTXLUHG�E\�WKH�PLG�VL]HG�HQWHUSULVH��7KLV�WHVW�LV�
important because, as hard as it may be to believe, there are CSPs who
intentionally take down their cloud service, sometimes for hours at a
time, to perform maintenance and install upgrades.
Of the eight providers that made it to this stage of testing, however, none failed. Amazon S3 and Microsoft Azure
ZHUH�WKH�WRS�WZR�SHUIRUPHUV�LQ�WHUPV�RI�DYHUDJH�SLQJ��EXW�GLIIHUHQFHV�DPRQJ�WKH�WRS�VL[�&63V�ZHUH�VPDOO��ZLWK�
WKH�H[FHSWLRQ�RI�1LUYDQL[��ZKRVH�DYHUDJH�SLQJ�ZDV�PRUH�WKUHH�WLPHV�WKDW�RI�WKH�IDVWHVW�&63��0LFURVRIW�$]XUH�
The purpose of this test is to determine whether CSPs can deliver the caliber of reliability required E\�WKH�PLG�VL]HG�enterprise.
Average Response Time
Please note that on this chart, shorter
bars represent faster speeds.
Page 8 / www.nasuni.com / info@nasuni.com
2011 © Nasuni Corporation, All Rights Reserved
+1.508.433.6200 / +1.800.208.3418 International US and Canada only
In terms of outages, Amazon
had the fewest with just 1.43
outages per month. However,
the duration of these outages
ZDV�QRW�VLJQL÷FDQW�DV�$PD]RQ�
had an uptime that essentially
equaled 100 percent. Availability
for Microsoft Azure, Peer1 and
Rackspace were comparable,
ZLWK�1LUYDQL[�DQG�$77�FRPLQJ�LQ�
ODVW�SODFH��1LUYDQL[��LQ�SDUWLFXODU��
H[SHULHQFHG�D�KLJK�QXPEHU�RI�
outages relative to this peer
group, though its availability was
�����SHUFHQW��$77�GLG�QRW�KDYH�D�
large number of outages, but their
duration impacted its availability,
which was the lowest of the top
VL[�SURYLGHUV�DW������SHUFHQW�
Provider Availability
Scalability)RU�WKLV�WHVW�1DVXQL�FRQWLQXRXVO\�ZULWHV�VPDOO�÷OHV���.%��ZLWK�PDQ\�
threads (concurrency) for several weeks or at least until it has written
����PLOOLRQ�REMHFWV��,GHDOO\��WKH�WHVW�ZRXOG�LQYROYH�ELOOLRQV�RI�÷OHV�RYHU�
a longer period of time, but the costs of performing such a test are
SURKLELWLYH��$V�LW�ZULWHV�WKHVH�÷OHV��1DVXQL�PHDVXUHV�DQG�WUDFNV�WKH�HUURU�
rates and performance. Two of the eight providers that went through this
stage of testing failed.
:LWKRXW�SURSHU�WHVWLQJ��LW�LV�LPSRVVLEOH�WR�GLIIHUHQWLDWH�EHWZHHQ�DQ�
LQGXVWULDO�VWUHQJWK�&63�DQG�D�OHVVHU�RSHUDWLRQ��,Q�IDFW��VRPH�SURYLGHUV�
,W�LV�QRW�GLI÷FXOW�WR�create something that looks like cloud storage. It is very GLI÷FXOW��KRZHYHU��WR�create a cloud that is WUXO\�VFDODEOH��UHOLDEOH�and always available.
Page 9 / www.nasuni.com / info@nasuni.com
2011 © Nasuni Corporation, All Rights Reserved
+1.508.433.6200 / +1.800.208.3418 International US and Canada only
have asked Nasuni to cease testing at this stage because they said it
was negatively impacting their customers, which is a truly frightening
statement. True cloud storage should be able to accommodate billions of
÷OHV�ZLWKRXW�DQ\�YLVLEOH�VWUDLQ���
Those CSPs that faced performance issues under Nasuni’s test are simply
not equipped to deliver an appropriate level of service to customers.
The top two performers were, yet again, Amazon S3 and Microsoft Azure.
For writing objects, the error rate for both was effectively zero. There
was a very small error rate with reading objects for Microsoft Azure at 0.07;
Amazon’s error rate for reading was effectively zero.
%HVW�RI�6KRZ��$PD]RQ�6��8OWLPDWHO\��VL[�&63V�SDVVHG�WKH�WHVW��
á� Amazon S3
á� AT&T Synaptic Storage as a Service
á� Microsoft Azure
CSPs’ Error Rates Under Scalability Testing
However, Amazon S3 and Microsoft Azure were clearly the standouts, with Amazon S3 offering the best overall
UHVXOWV��7KRXJK�1LUYDQL[�ZDV�IDVWHU�WKDQ�$PD]RQ�6��IRU�ODUJH�÷OHV��DQG�0LFURVRIW�$]XUH�ZDV�VOLJKWO\�IDVWHU�ZKHQ�
LW�FRPHV�WR�ZULWLQJ�÷OHV��QR�RWKHU�YHQGRU�SRVWHG�WKH�NLQG�RI�FRQVLVWHQWO\�IDVW�VHUYLFH�DFURVV�DOO�÷OH�W\SHV�DV�GLG�
$PD]RQ�6����$PD]RQ�6��KDG�WKH�IHZHVW�RXWDJHV�DQG�EHVW�XSWLPH��DQG�ZDV�WKH�RQO\�&63�WR�SRVW�D������HUURU�UDWH�
in both writing and reading objects. And though Microsoft Azure had a slightly faster average ping time than
Amazon S3 (likely because Amazon S3 is much more heavily used than Microsoft Azure), Amazon nevertheless
had the lowest variability.
Nasuni’s testing of CSPs continues, so these standings could change in future reports.
Analysis
Cloud storage has the potential to truly disrupt the storage market due to its DELOLW\�WR�SURYLGH�ÜLQKHUHQWÝ�GDWD�SURWHFWLRQ�DQG�XQOLPLWHG�VFDODELOLW\��ZKLOH�imposing minimal infrastructure and management requirements on customers.
Like most technologies, this nascent market can be broken into three segments: Large enterprise, midmarket and
VPDOO�EXVLQHVV�FRQVXPHU���0DQ\�ODUJH�HQWHUSULVHV�GR�QRW�ZDQW�WR�GHSHQG�RQ�DQ�RXWVLGH�YHQGRU�IRU�VWRUDJH�RI�FULWLFDO�
÷OHV�DQG��DV�VXFK��DUH�ZRUNLQJ�WR�EXLOG�WKHLU�RZQ�SULYDWH�FORXGV��1HYHUWKHOHVV��D�QXPEHU�RI�ODUJH�EXVLQHVVHV�WKDW�
IXQFWLRQ�SULPDULO\�RQOLQH��VXFK�DV�)DFHERRN�DQG�=\QJD��DUH�WDNLQJ�DGYDQWDJH�RI�&63V�IRU�WKHLU�VWRUDJH�LQIUDVWUXFWXUH��
á� Nirvanix
á� Peer1 Hosting
á� Rackspace Cloud
Page 10 / www.nasuni.com / info@nasuni.com
2011 © Nasuni Corporation, All Rights Reserved
+1.508.433.6200 / +1.800.208.3418 International US and Canada only
As a result, CSPs are largely oriented to serving these customers and their very large appetites for storage.
Midmarket enterprises, on the other hand, face a dilemma. Even if it were appropriate for the business to build a
private cloud (and in most cases, it is not), they are not large enough to possess the economies of scale they need
to build one, so they must rely on a CSP for cloud storage. However, as stated above, CSPs are largely focused
RQ�VHUYLQJ�YHU\�ODUJH�HQWHUSULVHV�DQG��ZLWKRXW�KHOS��PLG�VL]HG�HQWHUSULVHV�ZLOO�÷QG�FXVWRPHU�VHUYLFH�DW�WKH�
ODUJHU�FORXG�SURYLGHUV�ODFNLQJ��7KH�,7�GLUHFWRU�IRU�D�����HPSOR\HH�FRPSDQ\��IRU�H[DPSOH��ZLOO�÷QG�LW�GLI÷FXOW�WR�
UDLVH�VRPHRQH�RQ�WKH�SKRQH�DW�D�ODUJH�&63�WR�WURXEOHVKRRW�SUREOHPV���,Q�DGGLWLRQ��PLG�VL]HG�RUJDQL]DWLRQV�PD\�
GLVFRYHU�WKH\�ODFN�WKH�WHFKQLFDO�H[SHUWLVH�WR�LQWHUIDFH�SURSHUO\�ZLWK�WKH�&63ÚV�$3,V�DQG�RWKHU�DVSHFWV�RI�WKHLU�
WHFKQRORJ\��0DMRU�&63V�FDWHU�WR�GHYHORSHUV��QRW�,7�SHUVRQQHO�DW�HQG�XVHU�RUJDQL]DWLRQV�
Finally, there are a number of CSPs catering to the small business and consumer markets, but they are riddled
with security issues and do not provide robust data protection. Customers often cannot recover lost data and
VHUYLFH�OHYHO�DJUHHPHQWV�DUH�W\SLFDOO\�QRQ�H[LVWHQW��+RZHYHU��WKHVH�VKRUWFRPLQJV�PD\�EH�DFFHSWDEOH�WR�VRPH�LQ�
this segment of the market, since the data being sent to the cloud may not be terribly sensitive.
2I�WKH�WKUHH�PDUNHWV��WKH�PLGPDUNHW�IDFHV�WKH�PRVW�YH[LQJ�GLOHPPD��FRPSDQLHV�LQ�WKLV�VSDFH�UHTXLUH�WKH�
SHUIRUPDQFH�RI�WKH�WRS�&63V��EXW�WKH\�GR�QRW�GHOLYHU�WKH�YROXPH�RI�EXVLQHVV�WR�EH�ÜLQWHUHVWLQJÝ�WR�WKRVH�&63V��
0LG�VL]HG�HQWHUSULVHV�QHHG�WKH�KHOS�RI�SDUWQHUV�WKDW�FDQ�÷OO�WKLV�VHUYLFH�YRLG�E\�PDQDJLQJ�WKH�UHODWLRQVKLS�ZLWK�
WKH�&63��DQG�SURYLGLQJ�YDOXH�DGGHG�VHUYLFHV�VXFK�DV�HQKDQFHG�GDWD�SURWHFWLRQ��UHVSRQVLYH�FXVWRPHU�VHUYLFH�DQG�
guaranteed levels of performance.
The largest CSPs are the equivalent of the hard drive vendors in traditional storage, offering low prices and a high
TXDOLW\�UDZ�VWRUDJH�VHUYLFH�WKURXJK�VKHHU�RSHUDWLRQDO�H[FHOOHQFH�DQG�HFRQRPLHV�RI�VFDOH��%XW�UDZ�FORXG�VWRUDJH�
lacks performance, security, and any sort of comprehensive account management. Raw cloud storage is just as
unusable to most businesses as a solitary hard drive. In much the same way that the hard drive industry evolved,
the CSP market will continue to consolidate, leaving only a handful of players who have the massive economies
of scale required to compete effectively in what will eventually become a commodity business. And just as
companies such as EMC emerged to package commodity disk drives into intelligent systems, a new generation
of storage vendors is now emerging who are focused on harnessing the power of raw cloud storage to deliver
complete storage systems built to run businesses.
About Nasuni1DVXQL��D�QH[W�JHQHUDWLRQ�HQWHUSULVH�VWRUDJH�FRPSDQ\��GHOLYHUV�'DWD�&RQWLQXLW\�6HUYLFHV�WKDW�RIIHU�D�QHZ�EUHHG�
of data protection, accessibility and support to organizations that require their data be accessible 100 percent of
the time with no risk of data loss. Nasuni’s storage services offer primary data storage, offsite disaster recovery,
data backup and a 100 percent reliability guarantee that is supported by the storage industry’s most stringent
VHUYLFH�OHYHO�DJUHHPHQW��:LWK�1DVXQL��RUJDQL]DWLRQV�FDQ�VLPSOLI\�,7��HQVXUH�EXVLQHVV�FRQWLQXLW\��DQG�UHGXFH�WKH�
total cost of storage. For more information, visit ZZZ�QDVXQL�FRP.
Social Media Links
TwitterZZZ�WZLWWHU�FRP�QDVXQL
The Nasuni cloud storage blogKWWS���ZZZ�QDVXQL�FRP�EORJ
All company and product names are property of
their respective owners.
Page 11 / www.nasuni.com / info@nasuni.com
2011 © Nasuni Corporation, All Rights Reserved
+1.508.433.6200 / +1.800.208.3418 International US and Canada only
SOCS_BR_120111_MA
top related